This page intentionally left blank

Market-Valuation Methods in Life and Pension Insurance
In classical life insurance mathematics, the obligations of the insurance company
towards the policy holders were calculated on artificial conservative assumptions on
mortality and interest rates. However, the classical approach is being superseded
by developments in international accounting and solvency standards coupled with
theoretical advances in the understanding of the principles and methods for a more
market-based valuation of risk, i.e. its price if traded in a free market.
The book describes these new approaches, and is the first to explain them in con-
junction with more traditional methods. The exposition integrates methods and results
from financial and insurance mathematics, and is based on the entries in a life insurance
company’s market accounting scheme. With-profit insurance contracts are described
in a classical actuarial model with a deterministic interest rate and no investment
alternatives. The classical valuation based on conservative valuation assumptions is
explained and an alternative market-valuation approach is introduced and generalized
to stochastic interest rates and risky investment alternatives. The problem of incom-
pleteness in insurance markets is addressed using a variety of methods, for example
risk minimization, mean-variance hedging and utility optimization. The application
of mathematical finance to unit-linked life insurance is unified with the theory of
distribution of surplus in life and pension insurance. The final chapter provides an
introduction to interest rate derivatives and their use in life insurance.
The book will be of great interest and use to students and practitioners who need
an introduction to this area, and who seek a practical yet sound guide to life insurance
accounting and product development.
International Series on Actuarial Science
Mark Davis, Imperial College London
John Hylands, Standard Life
John McCutcheon, Heriot-Watt University
Ragnar Norberg, London School of Economics
H. Panjer, Waterloo University
Andrew Wilson, Watson Wyatt
The International Series on Actuarial Science, published by Cambridge University
Press in conjunction with the Institute of Actuaries and the Faculty of Actuaries, will
contain textbooks for students taking courses in or related to actuarial science, as
well as more advanced works designed for continuing professional development or
for describing and synthesising research. The series will be a vehicle for publishing
books that reflect changes and developments in the curriculum, that encourage the
introduction of courses on actuarial science in universities, and that show how actuarial
science can be used in all areas where there is long-term financial risk.
Market-Valuation Methods in Life and
Pension Insurance
THOMAS MØLLER
PFA Pension, Copenhagen
MOGENS STEFFENSEN
Institute for Mathematical Sciences, University of Copenhagen
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN-13 978-0-521-86877-8
ISBN-13 978-0-511-27035-2
© T. Møller and M. Steffenson 2007
2007
Information on this title: www.cambridge.org/9780521868778
This publication is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
ISBN-10 0-511-27035-6
ISBN-10 0-521-86877-7
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
hardback
eBook (NetLibrary)
eBook (NetLibrary)
hardback
Contents
Preface page ix
1 Introduction and life insurance practice 1
1.1 Introduction 1
1.2 The life insurance market 1
1.3 The policy holder’s account 4
1.4 Dividends and bonus 6
1.5 Unit-linked insurance and beyond 9
2 Technical reserves and market values 11
2.1 Introduction 11
2.2 The traditional composition of the liability 12
2.3 The market-based composition of the liability 18
2.4 The liabilities and principles for valuation 21
2.5 The liability and the payments 27
2.6 The surrender option 30
2.7 The free policy option 37
3 Interest rate theory in insurance 45
3.1 Introduction 45
3.2 Valuation by diversification revisited 46
3.3 Zero coupon bonds and interest rate theory 58
3.4 A numerical example 65
3.5 Bonds, interest and duration 71
3.6 On the estimation of forward rates 74
3.7 Arbitrage-free pricing in discrete time 78
3.8 Models for the spot rate in continuous time 93
3.9 Market values in insurance revisited 98
v
vi Contents
4 Bonus, binomial and Black–Scholes 101
4.1 Introduction 101
4.2 Discrete-time insurance model 103
4.3 The binomial model 115
4.4 The Black–Scholes model 124
4.5 Continuous-time insurance model 131
4.6 Generalizations of the models 143
5 Integrated actuarial and financial valuation 146
5.1 Introduction 146
5.2 Unit-linked insurance 148
5.3 The policy holder’s account 152
5.4 Hedging integrated risks under diversification 161
5.5 Hedging integrated risk in a one-period model 163
5.6 The multi-period model revisited 172
5.7 Hedging integrated risks 177
5.8 Traditional life insurance 199
6 Surplus-linked life insurance 200
6.1 Introduction 200
6.2 The insurance contract 203
6.3 Surplus 210
6.4 Surplus-linked dividends 215
6.5 Dividends linear in surplus 217
6.6 Bonus 222
6.7 Surplus- and bonus-linked dividends 226
6.8 Dividends linear in surplus and bonus 229
7 Interest rate derivatives in insurance 235
7.1 Introduction 235
7.2 Swaps and beyond in continuous time 236
7.3 Pricing of interest rate derivatives 240
7.4 Swaps and beyond in discrete time 246
7.5 A brief introduction to market models 254
7.6 Interest rate derivatives in insurance 257
7.7 A portfolio of contracts 257
Contents vii
Appendix 263
A.1 Some results from probability theory 263
A.2 The risk-minimizing strategy 266
A.3 Risk minimization for unit-linked contracts 267
A.4 Mean-variance hedging for unit-linked contracts 269
References 272
Index 278
Preface
Insurance mathematics and financial mathematics have converged during the
last few decades of the twentieth century and this convergence is expected
to continue in the future. New valuation methods are added to the traditional
valuation methods of insurance mathematics. Valuation and decision making
on the asset side and the liability side of the insurance companies are, to an
increasing extent, being considered as two sides of the same story.
The development has two consequences. Demands are made on practising
actuaries, whose education dates back to when financial mathematics was not
considered as an integrated part of insurance mathematics. By considering
the convergence as it applies to their daily work, such actuaries should be
kept abreast of this convergence. From this starting point, the ideas, concepts
and results of finance should be brought together to construct a path between
classical actuarial deterministic patterns of thinking and modern actuarial
mathematics. This is where stochastic processes are brought to the surface in
payment streams as well as in investment possibilities.
At the same time, present students of actuarial mathematics need to apply
financial mathematics to classical insurance valuation problems. These stu-
dents will typically, and should, meet financial mathematics in textbooks on
pure finance. However, to receive the full benefit of financial mathematical
skills, these skills need to be integrated and proven beneficial for classical
problems of insurance mathematics already on a student level.
International accounting standards have developed over the years. Denmark
has been at the forefront, implementing new accounting methods to replace
(assumed to be) conservative book values with real values based on market
information. Although the international accounting standards have not yet
been settled, the Danish approach to market valuation seems to be an important
step in the right direction. Many aspects of this approach are underpinned by
methods taken from mathematical finance.
ix
x Preface
The rationale for this book is that practising actuaries need an exposition
of financial methods and their applications to life insurance from the point
of view of a practitioner. Methods and applications are discussed in terms of
the Danish approach to market valuation. As a by-product, the book explains
to present students how financial methods known to them can be applied to
valuation problems in the life insurance market.
In 1995 and 1996, Tomas Björk and Ragnar Norberg gave courses in
financial mathematics and applications to life insurance at the University
of Copenhagen. These courses aroused our interest in the interplay between
finance and insurance. We studied the topics in our master theses, finished in
1996 (T.M.) and 1997 (M.S.), and continued our studies in our Ph.D. theses,
finished in 2000 and 2001, respectively.
Parts of the book (Chapter 2–5) are based on material which was developed
for a course on market valuation in life and pension insurance. This course was
organized by the Danish Actuarial Association in 2001. Each chapter was the
material for one course module and was written more or less independently of
the others. The material was originally written in Danish. In 2002 the material
was developed further and translated into English, and Chapter 1 was added.
In 2003, Chapters 6 and 7 were added on the occasion of The First Nordic
summer school in insurance mathematics, entitled “New Financial Products
in Insurance.” In 2004, the material was made consistent for notation and
terminology. However, it is still our intention that each chapter should be
readable more or less independently of the others. Therefore, some definitions
and introductions of quantities are repeated throughout the book.
The book suggests approaches to life insurance market-valuation problems.
The starting point is the version of with-profit insurance provided by Danish
life insurance companies. In order to help the reader follow the mathematical
description of this type of product, Chapter 1, written by Mogens Steffensen,
provides a non-mathematical introduction to life insurance practise in general.
In Chapter 2, also written by Mogens Steffensen, the with-profit insurance
contract is described at first in a classical actuarial model with a deterministic
interest rate and no investment alternatives. The classical valuation based on
conservative valuation assumptions is explained, and an alternative market-
valuation approach is introduced. Here, the partition of future payments in
guaranteed payments and non-guaranteed payments is important. Particular
attention is paid to the intervention options held by the policy holder, i.e.
the surrender and free policy (paid-up policy) options. Various approaches to
these options are suggested.
The market-valuation method introduced in Chapter 2 is generalized to a
stochastic interest rate in Chapter 3. Both discrete-time and continuous-time
Preface xi
bond market theory are introduced to a level such that the reader can follow
the reasoning behind replacing the discount factor in the market-valuation
formulas for guaranteed payments by zero coupon bond prices. Fundamental
financial concepts, such as arbitrage and market completeness, are introduced
in a bond market framework. Difference and differential equations for the
market value of guaranteed payments are derived. Chapter 3 is written by
Thomas Møller.
In Chapter 4 the market-valuation method introduced in Chapter 2 is gener-
alized to a situation with one risky investment alternative to the deterministic
interest rate. Both discrete-time and continuous-time stock market theory are
introduced to a level such that the reader can follow calculations of mar-
ket valuations of non-guaranteed payments in the case of two investment
alternatives. Fundamental financial concepts, such as arbitrage and market
completeness, are repeated in a stock market framework. Difference and dif-
ferential equations for the market value of the total payments, including the
non-guaranteed payments, are derived. Finally, the stock market is connected
to the stochastic bond market introduced in Chapter 3. Chapter 4 is written
by Mogens Steffensen.
The usual outline of introductory financial mathematics is to introduce the
fundamental financial concepts in a discrete and/or continuous stock market
model. Afterwards these are repeated in a discrete- and/or continuous-time
bond market model. In our exposition, the cut is different and is based on
entries in a life insurance company’s market accounting scheme. Valuating the
guaranteed payments, the important stochastic generalization of the classical
actuarial deterministic financial market lies at the introduction of a stochastic
bond market. A stochastic stock market comes into play when valuating the
non-guaranteed payments.
An alternative class of insurance contracts to with-profit insurance is unit-
linked insurance contracts, which are studied in Chapter 5, written by Thomas
Møller. This class of contracts and their market values are analyzed on the
basis of the stochastic financial markets introduced in Chapters 3 and 4.
The problem of genuine incompleteness in insurance markets is addressed.
The incompleteness in insurance markets stemming, for example, from mor-
tality risk is often taken care of in the literature by assuming risk neutrality of
the insurance company with respect to such a risk. Relaxing this assumption,
we suggest various approaches to incomplete market valuation. In particular,
valuation and optimal investment methods based on risk minimization, mean-
variance hedging and utility optimization are introduced and exemplified in
the case of unit-linked insurance.
xii Preface
InChapter 6, the applicationof mathematical finance tounit-linkedlife insur-
ance is unified with the theory of distribution of surplus in life and pension
insurance. The unification is based on a consideration of distribution of sur-
plus as an integrated part of the insurance contract. The notion of surplus and
various dividend and bonus schemes linked to this surplus are studied. In par-
ticular, explicit results are obtained in the case where dividends and bonus
payments are linear in the surplus. Chapter 6 is written by Mogens Steffensen.
Typically, insurance companies are faced by insurance liabilities that
extend up to sixty years into the future, whereas the financial markets typ-
ically do not offer bonds that extend more than thirty years into the future.
With market-based valuation methods, the value of both assets and liabili-
ties are affected by changes in the economic environment. Here, interest rate
derivatives seem to have become an important risk-management tool for life
insurance companies. An introduction to certain concepts and instruments
from the area of interest rate derivatives is therefore given in Chapter 7,
written by Thomas Møller. Examples are swap rates, swaps, floors, caps,
swaptions and CMS options. Various pricing methods are discussed, and it is
demonstrated how the financial impact on a life insurance company of these
instruments could be assessed.
The book studies approaches for market valuation of life insurance liabili-
ties. The various chapters address specific aspects of market-based valuation
and contain introductions to theoretical results from financial mathematics
and stochastic calculus that are necessary for the applications. A brief discus-
sion of the relation to existing books on financial mathematics and insurance
mathematics is given in the following list.
• Björk, T. (2004). Arbitrage Theory in Continuous Time, 2nd edn (Oxford:
Oxford University Press). Most of the theoretical results related to financial
mathematics presented here can be found in this book by Tomas Björk. To
some extent, the notation suggested by Björk is considered as “standard”
and is therefore used in the present book. Even the structural exposition of
certain topics of financial mathematics is inspired by Björk’s book.
• Briys, E. and de Varenne, F. (2001). Insurance: From Underwriting to
Derivatives: Asset Liability Management in Insurance Companies
(Chichester, UK: Wiley). The book discusses the convergence between the
insurance industry and the capital markets. It is less mathematical than the
current book and focuses on institutional aspects of the interplay between
the two fields. In contrast, the current book investigates the convergence of
the theories of financial mathematics and insurance mathematics and their
applications to market-based valuation.
Preface xiii
• Gerber, H.U. (1997). Life Insurance Mathematics (Berlin: Springer). This
provides an introduction to classical life insurance mathematics and can be
viewed as a necessary prerequisite for the current book. The concepts and
techniques discussed in Gerber’s book (for example, present values, specific
life insurance contracts, decrement series, Thiele’s differential equation)
are also used and explained in the present book. However, the current
manuscript has a completely different goal and goes considerably beyond
the introductory presentation in Gerber’s book.
• Koller, M. (2000). Stochastische Modelle in der Lebensversicherung (Berlin:
Springer). The book presents a framework where the underlying insurance
contracts are modeled by Markov chains and where stochastic interest rates
are allowed. The main difference between Koller’s book and the current
manuscript is that we focus more on market values and the application of
theories from financial mathematics in the area of life insurance; Koller’s
book deals more with the underlying Markov chains and on deriving dif-
ferential equations for the corresponding reserves (based on the work by
Hoem, Norberg and others).
We expect our readership to fall into two categories. Firstly, practising life
insurance actuaries who need an update of the mathematics of life insurance,
an introduction to financial mathematics in an insurance context and an
approach to market valuation in life insurance. Indeed, the book takes the
point of view of a practising actuary. Chapter 1 on life insurance practice will
provide the reader with sufficient insight into this practice.
Secondly, it is expected that the book will be read by students in actuar-
ial science, who have prerequisites in both life insurance mathematics and
mathematical finance, but want to see how these disciplines can and will be
combined in both theory and practice. By taking the viewpoint of a practising
actuary as a starting point, the student also sees how aspects of the classical
life insurance mathematics are implemented in practice.
The level is advanced. Basic knowledge of life insurance mathematics (such
as in Gerber’s book) is required. In addition, basic probability theory is required,
suchthat thereader canfollowtheintroductionof filtrations, martingales, stochas-
tic differential equations, etc. No previous knowledge of financial mathematics
is required. However, the theoretical results are at times developed quickly in
order to get to the applications, and at some points the reader would proba-
bly benefit from studying textbooks on financial mathematics for more details
and more background information; for example, Björk (2004).
We would like to thank Tomas Björk and Ragnar Norberg for arousing
our curiosity and for sharpening our understanding of the mathematics of life
xiv Preface
insurance and finance. The combination of basic knowledge in both areas and
a provoked curiosity made our studies of the interplay between these fields
possible, challenging and interesting. In addition, we wish to thank Ragnar
Norberg, Christian Hipp and Martin Schweizer for their guidance and support
during our Ph.D. studies. We would also like to thank Vibeke Thinggard
and Mikkel Jarbøl for valuable comments and discussions on earlier versions
of this material: in 2001 they were members of the Continued Professional
Development Committee under the Danish Actuarial Association and were
deeply involved in the organization of the course given in 2001.
1
Introduction and life insurance practice
1.1 Introduction
This chapter provides an introduction to life insurance practice with focus
on with-profit life insurance. The purpose is to give the reader sufficient
insight to benefit from the remaining chapters. In life insurance, one party,
the policy holder, exchanges a stream of payments with another party, the
insurance company. The exchanged streams of payments form, in a sense, the
basis of the insurance contract and the corresponding legal obligations. When
speaking of life insurance practice, we think of the way this exchange of
payments is handled and settled by the insurer. We take as our starting point
the idea of the policy holder’s account. This account can be interpreted as the
policy holder’s reserve in the insurance company and accumulates on the basis
of the so-called Thiele’s differential equation. Its formulation as a forward
differential equation plays a crucial role, and this chapter explains in words the
construction and the elements of this equation and its role in accounting. Note,
however, that the policy holder’s account is not in general a capital right held
by the insured but a key quantity in the insurer’s handling of his obligations.
1.2 The life insurance market
In this section we explain the most typical environments for negotiation and
contractual formulations for a life insurance policy. We distinguish between
defined benefits and what we choose to call defined contributions with partly
defined benefits.
Defined contributions with partly defined benefits cover the majority of
life insurance policies. The policy holder agrees with the insurance company
about a certain premium to cover a basket of benefits with, for example, a
1
2 Introduction and life insurance practice
life insurance sum paid out upon death before the termination of the contract
and/or a pension sum paid out upon survival until the termination of the
contract. The benefits agreed upon at issuance are set systematically low by
basing them on conservative assumptions on interest rates, insurance risk
and expenses. As the market interest rates, the market insurance risk and the
market costs evolve, a surplus emerges, and this surplus is to be paid back to
the policy holder. This typically happens by increasing one or more benefits.
This combination of known premiums reflected in guaranteed benefits, which
may be increased depending on the development in the market, categorizes
the contracts as defined contributions with partly defined benefits.
One construction is to increase all benefits proportionally such that the
ratio between, for example, the death sum and the pension sum is maintained.
Another construction is to keep the death sum constant or regulated with
some price index while residually increasing the pension sum. Only very
rarely is the surplus paid out in cash or used to decrease the premium instead
of increasing the benefits. Such rare constructions should in principle be
categorized as defined benefits with partly defined contributions since then a
basket of known benefits is combined with guaranteed premiums which may
be decreased depending on the development in the market.
The policies with defined contributions with partly defined benefits are
naturally classified as private, firm-based, or labor-based.
A private policy is agreed upon by a private person and the company.
The private person is the policy holder and negotiates the conditions in the
contract.
A firm-based policy is a contract which is part of an agreement between
an employer and an insurance company. The employer typically pays a part
of the premium but receives no benefits. The total premium paid is typically
a percentage of the salary. Although the terms of the contract are typically
negotiated between the employer and the company, the employees are still the
policy holders. The agreement between the employer and the company may
either be compulsory, in which case all employees are forced to participate,
or optional, in which case it is up to the employees to decide whether or not
to participate. Since the employer has no claims and no obligations besides
paying the premium, this premium can in many respects be interpreted as
salary.
A labor-based policy works in many respects as a firm-based policy except
for the fact that the employer and the employees are represented at the nego-
tiation by organizations rather than the employers and employees themselves.
These organizations typically take care of people employed in the uniformed
services or education. The result is an agreement where the employer is
1.2 The life insurance market 3
obliged to participate in the employees’ policy on terms agreed upon by the
organizations. The policy is then issued by a company taking care of all the
employees in a particular organization. Once the agreement has been made
and the contract has been issued, it works basically as a firm-based policy with
the employer and the employees as payers of premiums and the employees
as receivers of benefits. The total premium paid is typically a percentage of
the salary. The employees are the policy holders. The labor-based policy is
often part of a compulsory agreement, both the employer and the employed
are obliged for which to agree to minimum conditions. The premium part
paid by the employer can in many respects be interpreted as salary.
The defined benefit policies are usually a part of an agreement between
a firm and an insurance company, are therefore and comparable with the
firm-linked policy described above. The contract is negotiated indirectly by
settlement of the agreement. However, instead of sharing the premium defined
as a percentage of the salary between the employer and the employees, only
the employees’ part of the premium is defined as a percentage of the salary.
On the other hand, the benefits are also defined as a percentage of the salary.
This leaves a risk on the premium side. This risk is split between the employer
and the insurance company according to the agreement. If the risk is left to
the insurance company exclusively, then neither the employer nor the policy
holder participates in the development in the market but leaves all risk to the
insurance company. In contrast, if the risk is left to the employer exclusively,
the insurance company is pure administrator and takes no risk. As mentioned
above, a certain part of the defined contribution policies, where the surplus
is redistributed as cash or used to decrease the premiums, can actually be
considered as defined benefit policies with partly defined contributions.
The classification given above is fairly broad. When discussing details,
there may be a lot of differences in the concrete formulations of the various
agreements and contracts. Policies belonging to different classes may also
be mixed within an agreement and within a contract. In the following, we
concentrate on defined contributions with partly defined benefits. Although
most of the ideas presented in Chapters 2–5 may be applied to defined
benefits as well, all examples and interpretations take the defined contributions
with partly defined benefits policy as a starting point. as:eksist It should be
mentioned that in addition to the life insurance market described above, there
may exist a set of public insurance schemes. For instance, in Denmark the
national pension scheme is a pay-as-you-go scheme where present retirement
pensioners are covered by present tax payers. In addition, the Danish state
regulates a couple of particular funded pension schemes for people who
work.
4 Introduction and life insurance practice
1.3 The policy holder’s account
In our setting, the policy holder’s account or the technical reserve and its
dynamics are the technical elements in the handling and administration of
life insurance contract. We classify the changes of the technical reserve in
two different ways. The first way classifies the changes of income and outgo.
The other way classifies the changes in what was agreed beforehand in a
particular sense and additional changes made by the insurance company at
the discretion of the company.
The technical reserve can in certain respects be interpreted as a bank
account. The income on a bank account consists of capital injections and
capital return provided by the bank from capital gains on investment of
the account. The outgo on a bank account consists of capital withdrawals.
Correspondingly, the income on an insurance account consists of premiums
paid by the policy holder and return provided by the insurance company from
capital gains on investment of the account. And, correspondingly, the outgo
on an insurance account consists of the benefits paid to the policy holder.
However, two additional terms add to the change of the account. One term is
an outgo and covers the expense to the insurance company to administrate the
policy. Administration expenses on the bank account must also be paid, but
these are charged indirectly by a reduction of the return. The other additional
term which adds to the change of the account is the so-called risk premium,
which can be considered as an income or an outgo depending on its sign.
The risk premium is a premium that the policy holder pays from their
account; it may be positive, in which case it can be considered as an outgo, or
negative, in which case it can be considered as an income. The risk premium
is paid to cover the loss to the insurance company in case an insurance event
takes place in some small time interval. The amount of the potential loss is
also spoken of as the sum at risk. The premium for this coverage is set to the
expected value of this loss. Considering a so-called term insurance paying
out a sum upon death, the loss to the insurance company in case of death
equals the death sum which has to be paid out minus the technical reserve
which, on the other hand, can be cashed. The expected value of that loss
is the difference between the death sum and the technical reserve times the
probability of dying in some small time interval. The death sum exceeds the
technical reserve such that the risk premium is positive and can be considered
as an outgo. Considering instead a pure endowment insurance, the insurance
company cashes the technical reserve upon death and has no obligations.
The expected value of this gain is the technical reserve times the probability
of dying in a small interval. This results in a negative risk premium, and the
1.3 The policy holder’s account 5
risk premium may be considered as an income. In addition, the bank account
can be interpreted as an insurance contract where the technical reserve is
simply paid out upon death. This gives a potential loss upon death of the
technical reserve paid out minus the technical reserve cashed, whereby the
risk premium equals zero.
The term insurance and the pure endowment insurance are simple insurance
contracts. If we introduce such things as disability annuities, premium waiver
and deferred benefits, the picture becomes more blurred, but the underlying
idea is basically the same. Apart from the real incomes and outgoes in form of
premiums, returns and benefits, the policy holder pays or gains, depending on
the sign of the risk premium, for the risk imposed on the insurance company.
Another way of classifying the changes of the technical reserve is firstly to
identify the technical change which conforms with the guaranteed payments
and then identify the additional changes made by the insurance company at the
discretion of the company. When an insurance company issues an insurance
policy, it guarantees a minimum benefit which is based on a technical return.
Furthermore, the minimum benefit is based on a certain technical probability
of the insurance event, for example the probability of dying in a small time
interval. Finally, it is based on a technical amount for administration expenses.
Basically, it simply guarantees to pay out a benefit which is “fair” under a cer-
tain set of assumptions on return, insurance risk and expenses. However, this
set of assumptions is meant to be set so conservatively that a surplus emerges
over time. This surplus is provided by the policy holder due to conservatism
in assumptions and has to be paid back as the real market conditions evolve.
This is achieved by adding dividends to the policy holder’s account.
The law states that the surplus must be paid back to those who created it.
The usual way of allocating the dividends is to change the account, not in
correspondence with the technical assumptions, but in correspondence with
a set of assumptions that is more favorable to the policy holder. Then, we
can classify, element by element, the technical change and the additional
change. Concerning the return, the insurance company firstly pays the techni-
cal return. Secondly, it pays the difference between the more favorable return
and the technical return. Concerning the mortality, the insurance company
firstly collects a risk premium in correspondence with the technical probabil-
ity of death. Secondly, it pays back the difference between the risk premiums
in correspondence with the more favorable probability and the technical prob-
ability. Concerning the expenses, the insurance company firstly withdraws the
technical amount for expenses. Secondly, it pays back the difference between
the more favorable amount and the technical amount for expenses. The use
of the favorable assumption is that, element by element, the policy holder
6 Introduction and life insurance practice
should not be put into a worse position than if the technical assumption had
been used.
The favorable development of the account including dividends may be used
to reach a higher value to be paid out as a pension sum at the termination
of the contract. However, the policy holder may also wish that this favorable
development provides capital for an increase of benefits and/or a decrease of
premiums throughout the term of the policy. This has a feedback effect on the
dynamics of the account, since premiums, benefits and risk premiums need
to be adjusted in the light of such a change of payments. One construction
is to let the death sum and the pension sum increase proportionally, such
that the ratio between the two benefits is constant. However, the most typical
constructions are to let the death sum be constant or regulated by some price
index and then use the residual dividends to increase the pension sum.
We should mention an alternative application of dividends which has
gained ground in recent years. Instead of increasing benefits and/or decreasing
premiums, one may keep the guaranteed payments and instead change the
underlying technical assumptions. In this way, dividends are added to the
policy holder’s account without changing the guaranteed payments. One may
then ask: where did the money go and does allocation of dividends really
put the policy holder in a better position? The point is that paying out
dividends leads to what seems to be less favorable technical conditions.
However, the guaranteed payments are not changed and can therefore not
be less favorable. Furthermore, the consequence of less favorable technical
conditions is higher surplus contributions in the future. And since these surplus
contributions eventually have to be redistributed and reflected in payments,
the position is indeed favorable. By this construction, allocation of dividends
in a way postpone the increment of guaranteed benefits without postponing
the increment of the account.
1.4 Dividends and bonus
The premiums agreed upon at the time of issuance of an insurance policy
are “too high” compared with the benefits that are guaranteed at the time of
issuance. This disproportion is the source of surplus, and it is stated by law
that this surplus should be paid back to those policy holders who created it.
In practice, this happens in two steps. Firstly, the surplus is distributed among
the owners of the insurance company and the group of policy holders, and,
secondly, the surplus distributed to the policy holders is distributed among
the policy holders.
1.4 Dividends and bonus 7
So, why should the owners of the company take part in the surplus that
was created by policy holders? The problem is that “too high” may not be
high enough. The insurance company is allowed to invest not only in fixed
income assets, but also in stocks. Investment in stocks is, however, a risky
business, and the insurance company may end up in a situation where it is
not possible to increase the policy holder’s account by the technical interest
rate by means of capital gains. In that situation the owners of the insurance
company must still provide capital for the technical increments of the technical
reserve. Also, concerning mortality and expenses the insurance company may
experience a situation worse than that considered as the worst possible case
at the time of issuance. Concerning mortality and other kinds of insurance
risk, medical, sociological and demographic uncertainties play a different
role. The insurance company may need to help by injecting capital in the
technical reserve in order to live up to the technical conditions. The owners
of the insurance company must eventually cover the loss on the insurance
portfolio.
The risk that things may go wrong, leading to the owners having to
pay, is the reason why they, when things go right, deserve a share in the
surplus created by the insurance portfolio. However, the distribution of surplus
between owners and policy holders has to be fair in some sense. One of the
purposes of this book is to provide the insurance companies with tools and
ideas to make distributions that, to an increasing extent, are fair.
The part of the surplus distributed to the policy holders is deposited in
the so-called undistributed reserve. That is, this reserve is distributed to the
policy holders as a group but is not yet distributed among the policy holders.
The distribution among policy holders takes place by deciding on the favorable
set of assumptions introduced above. This mechanism transfers money from
the undistributed reserve to the individual policy holder’s account. As was
required from the distribution between owners and policy holders, the mutual
distribution between policy holders is also required to be fair. Fairness is
here given by the statement that the surplus should be redistributed to those
who earned it. A redistribution of the surplus to those who earned it has two
consequences.
The first consequence is that the insurance company is not allowed to
grow “large” undistributed reserves. This would systematically redistribute
surplus from the past and present policy holders to the future policy holders.
Thus, the insurance company needs to assign the undistributed reserve to the
individual technical reserves “in due time.” Here, “due time” is, of course,
closely connected to the risk of the insurance company owners to eventually
suffer a loss on the portfolio, which again connects to the owner’s share in
8 Introduction and life insurance practice
the surplus. If the undistributed reserve is high, then this reserve can take a
big loss before the owners have to take over. Then the insurance portfolio
pays a small, possibly zero, premium to the owners for taking risk. If the
undistributed reserve is low, then this reserve may easily run out, and the
insurance portfolio must pay a larger premium to the owners. This shows
that the solution to a fair distribution of the surplus between the owners and
the policy holders interacts substantially with the solution to a fair mutual
redistribution amongst policy holders over time.
The second consequence is that, given a redistribution to the present policy
holders, this must happen in a way that reflects which present policy holders
have contributed a lot to the surplus and which have contributed less. Such
a mechanism can be imposed by favorable assumptions on interest rates,
mortality and expenses. The return is proportional to the technical reserve,
the risk premium is proportional to the sum at risk, and the expense is typi-
cally formalized as a part of the premium. Therefore the individual technical
reserve, the sum at risk and the premium determine the individual share in
the total distribution. Once a redistribution from the undistributed reserve
among the policy holders is elected to happen now instead of later, the set of
favorable assumptions must to some extent reflect the present policy holder’s
contributions to the undistributed reserves.
Depending on the bonus scheme, the policy holder may experience the
redistribution in different ways. The typical construction is to increase the
benefits proportionally or to increase the pension sum residually, for example,
after a price index regulation of the death sum. The redistribution may also
be paid out as cash.
The redistribution of the surplus between the owners of the company and
the policy holders and the mutual redistribution between policy holders are
regulated by law and overseen by the supervisory authorities. Thus, they
are not directly specified in the contract. However, they make up a part of
the legislative environment in which the contract has been agreed upon, and
therefore they can be considered, in many respects, as part of the contract
itself. On the other hand, the conversion of dividends into payments on the
individual policy is a part of the individual policy conditions. Therefore,
this conversion is directly negotiable between the insurance company and the
policy holder or, in the case of a firm-linked or labor-linked contract, between
the company and the firm or labor organization, respectively. It is important
to realize how the legislative environment and the contract, in combination,
make up the conditions for all changes that are made over time by the
insurance company to the premiums and benefits agreed upon at issuance.
Firstly, the distribution between owners and policy holders (regulated by law);
1.5 Unit-linked insurance and beyond 9
secondly, the distribution mutually among policy holders (regulated by law);
and thirdly, by changing the terms in the individual policy (regulated by the
contract).
1.5 Unit-linked insurance and beyond
Several features characterize the participating policy as explained above. The
policy holder participates in a mutual fund, so to speak, together with the
other policy holders and together with the owners of the insurance company.
The legislative environment sets the conditions for this cooperation. However,
it may be difficult for the individual policy holder to understand whether the
conditions are followed, in particular concerning the several layers of fair
distributions. Even by representation of their ambassadors in the cooperation
in the form of the supervisory authorities, this may be a difficult task. One
way of avoiding the problems with fair distribution is that each and every
single policy holder forms their own individual fund. This is what happens
in unit-linked insurance.
In a unit-linked insurance contract, the policy holder does not participate
in a mutual fund but decides on their own investments to some extent. The
participating policies hold a very strong position in many countries and the
unit-linked market has been long in coming, but since the beginning of the
twenty-first century life insurance companies in these countries have started
to offer unit-linked insurance contracts. The unit-linked insurance contract
can be decorated with many different kinds of guarantees, and insurance
companies have shown some creativity on that point. However, the market
is still young, and there is still a lot of space for new developments and
improvements.
When giving up the investment cooperation and entering into unit-linked
contracts, policy holders typically also give up certain features of the par-
ticipating policy. By working with an undistributed reserve, one achieves a
smoothing effect of the market conditions. The undistributed reserve protects
the underlying technical reserves, and hereby the guaranteed payments, from
shocks in the market conditions. The technical reserves then only experience
a smoothed effect from such shocks. However, it is important to realize that
these smoothing effects do not rely particularly on the policy holders’ par-
ticipation in an investment cooperation. There is, in principle, no problem
in maintaining the smoothing effect in a unit-linked insurance policy. This
is only a matter of a proper definition of the unit to which the payments of
the contract are linked. Some insurance companies have introduced advanced
10 Introduction and life insurance practice
unit-linked products which maintain the smoothing effect. When speaking
of such products as unit-linked contracts, our characterization of unit-linked
products is that the investment game is an individual matter. If the investment
game is individualized, then the unit-linked contract stays unit-linked, no mat-
ter the complexity of the unit, even when including any kind of smoothing
effect.
One may argue that a unit-linked insurance contract endowed appropri-
ately with smoothing effects and guarantees is close, both in spirit and in
payments, to a participating policy. On the other hand, one may also argue
that it makes a huge difference whether the conditions for smoothing effects
and the guarantees are stated in the contract and individualized or are given in
the legislative environment by somewhat more vague statements on fairness.
One challenge is to incorporate the participating policies in an environment
of finance theory, as has successfully been achieved for unit-linked policies.
However, a proper description of unit-linked products in terms of finance
theory requires an enlargement of this environment. Furthermore, an appro-
priate enlargement of this environment is definitely needed to deal with the
complex nature of participating contracts and the special conditions of the life
and pension insurance market in general. One of the aims of the remaining
chapters of this book is to provide the reader with a box of tools that can be
applied for working with this challenge with theoretical substantiation.
2
Technical reserves and market values
2.1 Introduction
This chapter deals with some aspects of valuation in life and pension insurance
that are relevant for accounting at market values. The purpose of the chapter
is to demonstrate the retrospective accumulation of the technical reserve and
to formalize an approach to prospective market valuation. We explain and
discuss the principles underpinning this approach.
The exposition of the material distinguishes itself from scientific exposi-
tions of the same subject, see, for example, Norberg (2000) or Steffensen
(2001). By considering firstly the retrospective accumulation of technical
reserves, secondly the prospective approach to market valuation and thirdly
the underpinning principles, things are turned somewhat upside down here.
The aim is to meet the practical reader at a starting point with which he is
familiar.
The terms prospective and retrospective play an important role. The idea
of a liability as a retrospectively calculated quantity needs revision when
going from the traditional composition of the liability to a market-based
composition of the liability. This is an important step towards comprehending
both the market-valuation approach presented here and the generalization and
improvement hereof, taking into consideration more realistic actuarial and
financial modeling.
Throughout the chapter, we consider all calculations pertaining to the pri-
mary example of an insurance contract. This primary example is an endow-
ment insurance with premium intensity r, pension sum guaranteed at time
0, l
a
(0), and guaranteed death sum, l
ad
. Bonus is paid out by increasing the
pension sum. The insurance contract is issued at time 0 when the insured is x
years old and with a term of n years. Generalizations to other insurance con-
tracts are left to the reader. Throughout the chapter expenses are disregarded.
11
12 Technical reserves and market values
Assume that death occurs with a deterministic mortality intensity µ(s) at
age x +s, keeping in mind that this µ(s) of course depends on x then, and
assume that discounting is based on a deterministic interest rate r. An impor-
tant quantity is the discount factor from u ≥t to t, exp

u
t
r (s) ds

, which
we abbreviate by exp

u
t
r

. If the interest rate is constant, this discount
factor equals exp(−r (u−t)) =v
u−t
, where v =exp(−r). Another important
quantity is the survival probability from time t to u corresponding to age x+t
to age x +u, exp

u
t
µ(s) ds

, which we abbreviate by exp

u
t
µ

. The
actuarial notation for such a survival probability is
u−t
¡
x+t
.
We remind the reader about the following notation for the capital value at
time t of one unit of a pure endowment insurance, a temporary term insurance
and a temporary life annuity also used as a premium payment annuity:
n−t
E
x+t
=v
n−t
n−t
¡
x+t
=e

n
t
r+µ
.
A
1
x+tn−t
=

n
t
v
s−t
s−t
¡
x+t
µ(s) ds =

n
t
e

s
t
r+µ
µ(s) ds.
o
x+tn−t
=

n
t
v
s−t
s−t
¡
x+t
ds =

n
t
e

s
t
r+µ
ds.
In principle, the quantities o
x+t n−t
and A
1
x+t n−t
should be decorated with a bar,
but since only continuous-time versions appear in this chapter, this notation
is omitted. All quantities appear, on the other hand, with the decorations
∗ and o, for example
n−t
E

x+t
and o
o
x+t n−t
. These refer to the corresponding
fundamental quantities r and µ and the contents are obvious from the context.
2.2 The traditional composition of the liability
2.2.1 The technical reserve and the second order basis
In this section we study the technical reserve and how it arises from the second
order basis. The second order basis is a pair

r
o
. µ
o

containing the second
order interest rate and the second order mortality rate by which the technical
reserve accumulates.
We consider an accumulation of the technical reserve by the second order
basis. This corresponds to a difference equation with initial condition as
follows:
AV

(t) =r
o
(t) V

(t) At +rAt −µ
o
(t) R

(t) At. (2.1)
V

(0) =0.
2.2 Traditional composition of the liability 13
where R

is the sum at risk given by
R

(t) =l
ad
−V

(t) .
and where At is the time unit chosen for the accumulation. Actually, Equa-
tion (2.1) is a discrete-time version of a corresponding differential equation
which can be obtained by dividing the difference equation by At and letting
At go to zero. The corresponding differential equation with initial condition
reads as follows:
d
dt
V

(t) =r
o
(t) V

(t) +r−µ
o
(t) R

(t) . (2.2)
V

(0) =0.
This differential equation can be solved over the time interval (0. t] and
the initial condition leads to the retrospective form:
V

(t) =

t
0
e

t
s
r
o

o

r−l
ad
µ
o
(s)

ds. (2.3)
The differential equation, Equation (2.2), solved over the time interval (t. n]
leads to the prospective form,
V

(t) =l
ad
A
1o
x+t n−t
+V

(n)
n−t
E
o
x+t
−ro
o
x+tn−t
. (2.4)
where V

(n) is the terminal value of the technical reserve.
The second order basis is a decision variable held by the insurer that is to
be chosen within certain legislative constraints and market conditions. One
notes that in the prospective form, Equation (2.4), the second order basis
over (t. n] appears together with the terminal value of the technical reserve
V

(n). Since these may be unknown at time t, Equation (2.4) is a repre-
sentation of the solution to Equation (2.2), but not a constructive tool for
calculation of the technical reserve. Nevertheless, if the terminal value V

(n)
is interpreted as a terminal benefit, the prospective form expresses the tech-
nical reserve as a prospective value of all payments valuated under the future
second order basis.
2.2.2 The technical reserve and the first order basis
In this section we study the technical reserve and how it arises from the first
order basis. The first order basis is a pair (r

. µ

) containing the first order
interest rate and the first order mortality rate under which the guaranteed
benefits are set according to the equivalence principle.
14 Technical reserves and market values
We consider an accumulation of the technical reserve by the first order
basis. This corresponds to a differential equation with initial condition as
follows:
d
dt
V

(t) =r

(t) V

(t) +r−µ

(t) R

(t) +o(t) . (2.5)
V

(0) =0.
where the dividend rate o is given by
o(t) =

r
o
(t) −r

(t)

V

(t) +

µ

(t) −µ
o
(t)

R

(t) . (2.6)
The rate of dividends is determined such that the technical reserve in Sec-
tion 2.2.1 coincides with the technical reserve in this section. Given a second
order basis the rate of dividends is determined by Equation (2.6). On the
other hand, given a rate of dividends o, any pair

r
o
. µ
o

conforming with
Equation (2.6) is a candidate for the second order basis.
The differential equation, Equation (2.5), solved over |0. t] leads to the
retrospective form,
V

(t) =

t
0
e

t
s
r


r−l
ad
µ

(s) +o(s)

ds. (2.7)
On the basis of the technical reserve, the pension sum guaranteed at time t is
calculated in accordance with the equivalence principle,
l
a
(t) =
V

(t) +ro

x+t n−t
−l
ad
A
1∗
x+t n−t
n−t
E

x+t
. (2.8)
Hereafter, we can write the prospective form as follows:
V

(t) =l
ad
A
1∗
x+t n−t
+l
a
(t)
n−t
E

x+t
−ro

x+t n−t
. (2.9)
Note that from Equation (2.8) we get l
a
(n) =V

(n) such that the technical
reserve at the terminal time n coincides with the pension sum at that time.
This motivates the interpretation of the technical reserve V

(n) as the terminal
benefit at the end of the Section 2.2.1. In the following, we write l
a
(n) instead
of V

(n) when it is appropriate to think of V

(n) as the pension sum. Note
that Equation (2.8) also sets the guaranteed pension sum at time 0:
l
a
(0) =
ro

xn
−l
ad
A
1∗
xn
n
E

x
.
Since l
a
(t) is calculated on the basis of the retrospectively derived
technical reserve, we see that Equation (2.9) is a representation of V

(t)
but not a constructive tool for its derivation. Nevertheless, the prospective
2.2 Traditional composition of the liability 15
form expresses the technical reserve as a prospective value of the payments
guaranteed at time t valuated under the first order basis.
The technical reserve at time t covers payments guaranteed at time t. Later,
we use a technical reserve at time u ≥ t for payments guaranteed at time t.
We choose to introduce this quantity now and denote it by V

(t. u). Then
V

(t. u) =l
ad
A
1∗
x+un−u
+l
a
(t)
n−u
E

x+u
−ro

x+un−u
.
and for u ≥t we have a differential equation with initial and terminal condi-
tions, respectively, as follows:
o
ou
V

(t. u) =r

(u) V

(t. u) +r−µ

(u) R

(t. u) . (2.10)
V

(t. t) =V

(t) .
V

(t. n) =l
o
(t) .
where R

(t. u) is the sum at risk, given by
R

(t. u) =l
ad
−V

(t. u) .
2.2.3 The undistributed reserve and the real basis
In this section we study the undistributed reserve and discuss how the undis-
tributed reserve arises from the real basis. The real or third order basis is
the pair (r. µ) containing the real or third order interest rate and the real or
third order mortality rate by which the total reserve accumulates.
We consider an accumulation of the total reserve by the real basis. This
corresponds to the differential equation with initial condition as follows:
d
dt
U (t) =r (t) U (t) +r−µ(t) R(t) . (2.11)
U (0) =0.
where R is the sum at risk, given by
R(t) =l
ad
−U (t) .
The differential equation, Equation (2.11), solved over |0. t] leads to the
retrospective form:
U (t) =

t
0
e

t
s
r+µ

r−l
ad
µ(s)

ds. (2.12)
16 Technical reserves and market values
The differential equation, Equation (2.11), solved over (t. n] leads to the
prospective form:
U (t) =l
ad
A
1
x+t n−t
+U (n)
n−t
E
x+t
−ro
x+t n−t
. (2.13)
where U (n) is the terminal value of the total reserve.
The undistributed reserve X is calculated residually as the difference
between the total reserve and the technical reserve:
X(t) =U (t) −V

(t) . (2.14)
By differentiation, we obtain the retrospective form:
X(t) =

t
0
e

t
s
r+µ
(c (s) −o(s)) ds. (2.15)
where the contribution rate c is defined by
c (t) =(r (t) −r

(t)) V

(t) +(µ

(t) −µ(t)) R

(t) . (2.16)
The retrospective form given in Equation (2.15) shows how the undistributed
reserve consists of past contributions minus past dividends.
Now we put up the following condition on the second order basis:
X(n) =U (n) −V

(n) =0. (2.17)
Under this condition we have the following prospective form for the undis-
tributed reserve:
X(t) =

n
t
e

s
t
r+µ
(o(s) −c (s)) ds. (2.18)
This can be verified by differentiation of Equations (2.15) and (2.18) with
respect to t. Equation (2.18) shows how the undistributed reserve is consumed
in the future by letting the dividend rate differ from the contribution rate.
Furthermore, with V (t) defined as the prospective value of future payments
as follows:
V (t) ≡l
ad
A
1
x+t n−t
+l
a
(n)
n−t
E
x+t
−ro
x+t n−t
. (2.19)
the condition on the second order basis, Equation (2.17), implies that
U (t) =V (t) . (2.20)
This is obtained by replacing U (n) in Equation (2.13) by V

(n) =l
a
(n). In
this chapter, we work under the assumption from Equation (2.17) such that
Equation (2.20) holds. We therefore switch between U and V, in accordance
with Equation (2.20), from situation to situation depending on whether it is
beneficial to think of the quantity as retrospective or prospective.
2.2 Traditional composition of the liability 17
Note that the future second order basis appears in Equations (2.18) and
(2.19) through dividends and the terminal pension sum, respectively. This
means that Equations (2.18) and (2.19) are representations, following from
the condition X(n) =0, of the total and the undistributed reserves but are not
constructive tools for their derivation. Nevertheless, these forms express the
total and the undistributed reserves as prospective values of different future
payments depending on the future second order basis valuated under the real
basis. The total reserve is a value of the total payments and the undistributed
reserve is a value of dividends minus contributions.
Equations (2.12) and (2.15) are definitions of U and X whereas Equa-
tions (2.18) and (2.20) are representations of the same quantities based on
the condition X(n) = 0. However, one may also start out with these repre-
sentations as definitions, hereby including the future second order basis in
the definition of U and X. If we then restrict the second order basis to obey
X(n) =0 with X defined by Equation (2.15), the formulas above show that
further specification of the future second order basis is redundant. The quanti-
ties are hereafter given simply by the retrospective formulas, Equations (2.12)
and (2.15).
The condition U (n) −l
a
(n) = 0 can, by multiplication of e

n
0
r+µ
, be
written as follows:
l
a
(n)
n
E
x
+l
ad
A
1
xn
−ro
xn
=0.
This shows that the condition X(n) =0 is the same as performing the equiv-
alence principle on the total payments under the real basis.
It is important to understand that it is the condition X(n) =0 that spares
us from discussing the future second order basis further. If, for one reason or
another, the condition X is not to be fulfilled, the future second order basis
is inevitably brought into the quantity V.
Traditionally, one would, in addition to the terminal condition X(n) =0,
impose a solvency condition in the following form:
X(t) ≥0. 0 ≤t ≤n. (2.21)
In practice, the undistributed reserve must, together with other sources of
capital, excess a value based on the technical reserve, the sum at risk and the
premium. The purpose of Equation (2.21) is to secure that U at any point
in time covers the technical reserve interpreted as the value of guaranteed
payments valuated by the first order basis, including the safety margins this
may contain. Thus, in the traditional composition of the liability, one would
not accept that U ≤ V

. One of the potentials of accounting at market value
is to set up alternative solvency rules in terms of market values. As we shall
18 Technical reserves and market values
see, solvency rules such as U ≥ V

are, in principle, not necessary in the
market-based composition of the liabilities.
2.3 The market-based composition of the liability
2.3.1 Guaranteed payments and the market reserve
In this section we introduce the market reserve for the guaranteed payments.
The guaranteed payments at time t are given by the premium rate r, the
death sum l
ad
and the pension sum l
a
(t). The market reserve at time t is
given by the prospective formula,
V
g
(t) =l
ad
A
1
x+t n−t
+l
a
(t)
n−t
E
x+t
−ro
x+t n−t
. (2.22)
Thus, V
g
(t) is a prospective value of the guaranteed payments under the real
basis.
The market reserve at time t, V
g
(t), covers payments guaranteed at time
t. Later, we use a market reserve at time u ≥t for the payments guaranteed at
time t. We choose to introduce this quantity now and denote it by V
g
(t. u).
Then
V
g
(t. u) =l
ad
A
1
x+un−u
+l
a
(t)
n−u
E
x+u
−ro
x+un−u
.
and for u ≥t we have a differential equation with initial and terminal condi-
tions, respectively, as follows:
d
du
V
g
(t. u) =r (u) V
g
(t. u) +r−µ(u) R
g
(t. u) . (2.23)
V
g
(t. t) =V
g
(t) .
V
g
(t. n) =l
a
(t) .
where R
g
(t. u) is the sum at risk, given by
R
g
(t. u) =l
ad
−V
g
(t. u) .
2.3.2 Bonus payments and the bonus potential
In this section we introduce the bonus payments, the bonus potential, the
individual bonus potential and the collective bonus potential. In Section 2.2.3
we introduced U and X and we concluded that the retrospective quantities
were constructive tools for the calculation of the prospective quantities given
by Equations (2.18) and (2.20), given the condition on the second order basis,
X(n) = 0. Also in this section we define prospective values based on the
2.3 Market-based composition of the liability 19
future payments. As we shall see, the condition X(n) = 0, with X given by
Equation (2.15), again leads to retrospective calculation formulas.
The bonus payments at time t are the payments which are not guaranteed
at time t. The guaranteed payments at time t are given by the premium r,
the death sum l
ad
and the pension sum l
a
(t). The real payments differ from
these payments by the pension sum l
a
(n) only. The difference is exactly
the pension sum l
a
(n) −l
a
(t) making up the bonus payments. This bonus
payment has the following market value:
V
b
(t) =(l
a
(n) −l
a
(t))
n−t
E
x+t
. (2.24)
Thus, V
b
(t) is a prospective value under the real basis of the payments not
guaranteed. We also speak of V
b
(t) as the bonus potential at time t.
The bonus potential at time t, V
b
(t), covers payments not guaranteed at
time t. Later, we use a market-based reserve at time u ≥ t for the payments
not guaranteed at time t. We choose to introduce this quantity now and denote
this by V
b
(t. u). Then
V
b
(t. u) =(l
a
(n) −l
a
(t))
n−u
E
x+u
.
and for u ≥t we have a differential equation with initial and terminal condi-
tions, respectively, as follows:
o
ou
V
b
(t. u) =(r (u) +µ(u)) V
b
(t. u) .
V
b
(t. t) =V
b
(t) . (2.25)
V
b
(t. n) =l
o
(n) −l
o
(t) .
Now, we include the condition on the second order basis that X(n) = 0
with X given by Equation (2.15). By means of the differential equations (2.11)
and (2.23), it can now be shown that V
b
(t. u) =U (u)−V
g
(t. u), from which
it follows that
V
b
(t) =U (t) −V
g
(t) . (2.26)
We note that the bonus potential at time 0 equals the negative market reserve,
−V
g
(0), since U (0) =0.
In general, the quantity V
b
defined by Equation (2.24) depends on the
future second order basis through the terminal pension sum. However, as in
Section 2.2.3, the condition X(n) = 0 saves us from further specifications
since we have the representation in Equation (2.26).
It turns out to be informative to decompose the bonus potential into
two reserves: the individual bonus potential V
ib
and the collective bonus
potential V
cb
.
20 Technical reserves and market values
We consider the situation
V ≥V

≥V
g
. (2.27)
Then V
b
is decomposed as follows:
V
b
=V
cb
+V
ib
=(V −V

) +(V

−V
g
) .
Firstly, consider the individual bonus potential. From the differential equa-
tions (2.10) and (2.23) and the terminal condition V

(n. t) −V
g
(n. t) =
l
a
(t) −l
a
(t) =0, one obtains the prospective form,
V
ib
(t) =V

(t) −V
g
(t) =

n
t
e

s
t
r+µ
c (t. s) ds. (2.28)
where
c (t. s) =(r (s) −r

(s)) V

(t. s) +(µ

(s) −µ(s)) R

(t. s) . (2.29)
Thus, the individual bonus potential is simply the market value of the safety
margins in the guaranteed payments. This explains the term individual bonus
potential. We note that the individual bonus potential at time 0 equals the
negative market reserve, −V
g
(0), since V

(0) =0.
Secondly, we consider the collective bonus potential,
V
cb
(t) =V (t) −V

(t) =X(t) . (2.30)
The collective bonus potential can also be calculated residually as the bonus
potential minus the individual bonus potential. This explains the term collec-
tive bonus potential. We note that the collective bonus potential at time 0
equals zero since U (0) =V

(0) =0.
We now turn to the general case where Equation (2.27) does not necessarily
hold. If V

- V
g
, we set the individual bonus potential to zero using the
following formula:
V
ib
=(V

−V
g
)
+
=max(V

. V
g
) −V
g
. (2.31)
where (o)
+
equals o if o ≥0 and 0 if o -0. This is equivalent to
V
ib
(t) =

n
t
e

s
t
r+µ
c (t. s) ds

+
. (2.32)
In Sections 2.5.1 and 2.6.3, we discuss alternative definitions of the individual
bonus potential.
If we do not have the solvency constraint X(t) ≥0, it may also happen that
V -V

. In that case we set the collective bonus potential to zero. Taking into
2.4 Liabilities and principles for valuation 21
account all possible relations between V, V

and V
g
, the collective and the
individual bonus potential are formalized by the following general formulas:
V
cb
=

V −V
g
−(V

−V
g
)
+

+
.
V
ib
=V −V
g

V −V
g
−(V

−V
g
)
+

+
.
From these formulas we easily see that V
cb
and V
ib
are positive and sum
up V
b
.
2.4 The liabilities and principles for valuation
In this section we interpret the liabilities put up in the previous sections.
We represent these liabilities as conditional expected values and discuss the
principles underpinning this representation.
In Section 2.2 we presented formulas of both retrospective (Equations (2.3),
(2.7), (2.12), (2.15)) and prospective (Equations (2.4), (2.9), (2.18), (2.20))
type, and in Section 2.3 we presented primarily prospective formulas (Equa-
tions (2.22), (2.24), (2.28), (2.30)). We now discuss the elements in these
prospective formulas. This enables us to generalize these formulas to other
types of insurance. This also helps us in the search for theoretically substan-
tiated methods for calculation of values in the case where we do not require
from the second order basis that X(n) =0.
We remind the reader that a liability is a value set aside by the insurer in
order to be able to meet certain obligations in the future. Elementary examples
of such payments are the benefits of a pure endowment insurance, a temporary
term insurance or a temporary life annuity. These benefits appear as building
blocks in a number of insurance types, for example the level premium paid
endowment insurance, which is the main example given in this chapter.
Assume that Z is a process measuring whether death has occurred. A
process is here defined as a continuum of stochastic variables, indexed by
time, such that Z at time t assumes a certain value Z(t). Then
Z(t) =
¸
0. if death has not occurred at time t.
1. if death has occurred at time t.
(2.33)
Such a process, assuming the values 0 or 1, is called an indicator process for
the condition which has to be fulfilled for the process to assume the value 1.
Thus, Z is an indicator process for the insured to be dead.
We introduce another process N counting the number of deaths, i.e.
N (t) = number of deaths until time t, and in this situation, of course, Z =N.
22 Technical reserves and market values
0
alive →
1
dead
Figure 2.1. Survival model.
The process I is defined by I (t) = 1 −Z(t) = 1 −N (t), and I is thus an
indicator process for the insured to be alive. The process Z is illustrated in
Figure 2.1. The underlying stochastic variable, determining all the stochastic
processes Z, N and I at any point in time is the remaining lifetime at time 0
corresponding to age x, which we denote by T
x
.
Now the elements of the prospective formulas can be written as follows:
n−t
E
x+t
=E
t
¸
e

n
t
r
I (n)
¸
.
o
x+t n−t
=E
t
¸

n
t
e

s
t
r
I (s) ds
¸
.
A
1
x+t n−t
=E
t
¸

n
t
e

s
t
r
dN (s)
¸
.
(2.34)
where E
t
denotes an expectation conditional on Z(t) =0. We now interpret
the elements in the square brackets in Equations (2.34).
n−t
E
x+t
. This is the conditional expectation of the discounted benefit I (n).
o
x+t n−t
. The integral is interpreted as follows:

n
t
e

s
t
r
I (s) ds =I (t)

min(T
x
.n)
t
e

s
t
r
ds. (2.35)
On the left hand side of Equation (2.35) the value is written as the
discounted benefits since I (s) ds is exactly the benefit over the time
interval (s. s +ds].
A
1
x+t n−t
. The quantity dN (s) is the change in N at time s, such that
dN (s) = 0 if s = T
x
and dN (s) = 1 if s = T
x
. If the insured survives
until time n, dN (s) = 0 for t ≤ s ≤ n. Thus, dN (s) is the benefit at
time s from one unit of a term insurance. The integral now becomes a
sum of infinitely many zeros plus one discount factor from T
x
to t if
death occurs before time n, i.e.

n
t
e

s
t
r
dN (s) =I (t) (1−I (n)) e

T
x
t
r
.
The fundamental elements in Equations (2.34) are discounting (e

s
t
r
),
conditional expectation (E
0
t
|·]) and the elementary payments (I (n) . I (s) ds
and dN (s)). Whereas the elementary payments are just what we want to
2.4 Liabilities and principles for valuation 23
valuate, the other elements, discounting and the conditional expectation are
based on some principal considerations about what a value is and how it is
determined.
2.4.1 Absence of arbitrage
In this section we introduce the principle of no arbitrage and discuss how
discounting connects to this principle.
We assume that payments to the insurer are deposited in a bank account
with continuously accrued interest. Negative payments in the form of benefits
are withdrawn from the bank account. We denote by S
0
(t) the value at time
t of one unit deposited in the account at time 0, and we find that the value
at time t of one unit deposited at time s equals S
0
(t) ¡S
0
(s). We assume that
there exists a short rate of interest r such that
d
dt
S
0
(t) =r (t) S
0
(t) . S
0
(0) =1. (2.36)
i.e.
S
0
(t) =e

t
0
r
.
Instead of rushing to insert a corresponding discount factor appropri-
ately, we hesitate and discuss the real content of S
0
. What we are really
doing here is mathematical finance dealing with arbitrage-free prices and
investment strategies in a particular financial market. By introducing S
0
we have started the specification of this market. We have specified an
asset in which the insurer can invest. On the other hand, we have not
specified any investment alternatives. Hereby, we have specified a finan-
cial market containing only one asset S
0
in which the insurer invests all
payments.
If S
0
is deterministic, i.e. if r is deterministic, we can easily valuate
deterministic payments, and bond prices and prices of options on bonds etc.
are easily found. A zero coupon bond is an asset which pays one unit at
the terminal time n, and we denote by P (t. n) its value at time t. If S
0
is
deterministic, we have that
P (t. n) =e

n
t
r
.
Even if this may seem obvious, it is a good idea to underpin the result by a
so-called arbitrage argument as follows. Assume that P (t. n) =exp

n
t
r

+
a. Now assume that we sell the bond and invest its price, P (t. n), in S
0
, i.e.
we buy P (t. n) ¡S
0
(t) units of the asset S
0
. At time n the value of our
24 Technical reserves and market values
investments, after fulfillment of the obligation to pay one unit to the owner
of the bond, is given by
P (t. n) S
0
(n) ¡S
0
(t) −1 =ae

n
t
r
.
If a =0 this value is created without any risk. We do not accept the possibility
of creating value without risk and conclude that a =0.
Thus, an arbitrage argument deals with prices and investment strategies
avoiding the possibility of risk-free capital gains beyond the interest rate. This
is the principle of no arbitrage.
If S
0
is not deterministic, it is by no means clear what can be said about
P (t. n). Certain areas of mathematical finance deal with questions such as:
Given a stochastic model for r, what can be said about P (t. n) in general
and about its relation to P (t. u) . t - u - n, in particular? Such considera-
tions typically take the principle of no arbitrage as the theoretical starting
point.
Throughout this chapter r is assumed to be deterministic. Chapter 3 deals
with stochastic interest rates.
2.4.2 Diversification
In this section we introduce the principle of diversification and discuss how
conditional expectation connects to this principle.
In Section 2.4.1 we stated that valuation of deterministic payments is
simple if S
0
(t) is assumed to be deterministic. If, for example, the payment
I (n), which is the benefit at time n, is known at time 0, we simply obtain
the value
e

n
0
r
I (n) .
However, the time of death and other conditions which may determine a
payment in general are not known at time 0. Thus, exp

n
0
r

I (n) cannot
be interpreted as a value but as a stochastic present value. The question is,
what can be said about the value of a stochastic present value?
A special situation arises if the insurer issues, or can issue, contracts to
a “large” number m of insured with independent and identically distributed
payments. Denote by I
i
(t) the function indicating that insured number i
is alive at time t. Then the law of large numbers comes into force and
2.4 Liabilities and principles for valuation 25
we conclude that the total present value per insured converges towards the
expected present value of a single contract as the number of contracts is
increased, i.e.
1
m
m
¸
i=1
e

n
0
r
I
i
(n) →e

n
0
r
E|I (n)] as m→.
This result can be generalized to a situation when the insurance contracts do
not have the same terminal time n. If death is assumed to occur with intensity
µ, we obtain the following candidate to the value of I (n) at time 0:
e

n
0
r
E|I (n)] =e

n
0
r+µ
=
n
E
x
. (2.37)
giving the liability at time 0 of a pure endowment insurance issued to an
insured who at time 0 is x years old. We say that the value in Equation (2.37)
builds on the principles of no arbitrage and diversification.
It is important to note that none of the presented principles, no arbitrage
and diversification, solve the valuation problem by itself. We mention that the
mathematical financial term for the combination of the principles is the princi-
ple of no asymptotic arbitrage, and particular asymptotic arbitrage arguments
lead exactly to the value given in Equation (2.37).
The value in Equation (2.37) is a good candidate for the value at time 0 of
I (n) paid at time n. For various reasons, one is interested in evaluating future
payments at any point in time during the term of the contract. If one wishes
to sell the obligation of the payments, one needs to set a price. But even if
one does not want to sell, various institutions may be interested in the value
of future payments. The owners of the insurance company and other investors
are interested in the value of the payments for the purpose of assessing the
value of the insurance company itself; regulatory authorities are interested in
securing that the insurer can meet these payments with a large probability and
put up solvency rules which are to be met; the tax authorities are interested
in the surplus of the insurer as a basis for taxation. All these institutions are
interested in the value of future obligations.
By repeating the reasoning leading to the value
n
E
x
, we can put up the
liability at time t. Assuming that the insurer knows that the insured is alive
at time t, we obtain the following value:
e

n
t
r
E
t
|I (n)] =e

n
t
r+µ
=
n−t
E
x+t
.
giving the liability at time t of an insurance contract issued to an insured who
was x years old at time 0 and who is alive at time t.
26 Technical reserves and market values
2.4.3 The market-based liability revisited
In this section we show how the principles of no arbitrage and diversification
underpin the market-based composition of the liability.
By a simple rewriting of the prospective market-based liabilities, we unveil
the principles on which these are based. From Equations (2.22), (2.24), (2.28)
and (2.30) we obtain with help from Equation (2.34), the following:
V
g
(t) = E
t
¸

n
t
e

s
t
r

l
ad
dN (s) −I (s) rds

+ l
a
(t) e

n
t
r
I (n)
¸
. (2.38)
V
b
(t) = E
t
¸
(l
a
(n) −l
a
(t)) e

n
t
r
I (n)
¸
. (2.39)
V

(t) −V
g
(t) = E
t
¸

n
t
e

s
t
r
I (s) c (t. s) ds
¸
. (2.40)
V (t) −V

(t) = E
t
¸

n
t
e

s
t
r
(o(s) −c (s)) I (s) ds
¸
. (2.41)
All these quantities are seen to build on the principles of no arbitrage and
diversification. The principles leave tracks in terms of discount factors and
conditional expectations, respectively. We recall the interpretation of the
elements. Compared with the elementary payments in Equation (2.34) we see
that the death sum l
ad
, the pension sum l
a
(t) and the premium r appear in
V
g
(t). The bonus payment l
a
(n) −l
a
(t) appears in V
b
(t). In V

(t) −V
g
(t)
the payment process can be interpreted as continuous payments of a life
annuity with time dependent annuity rates where the safety margin c (t. s)
makes up the annuity rate at time s ≥ t. The rate (o(s) −c (s)) plays a
corresponding role in the quantity V (t) −V

(t).
Now consider the prospective versions of the technical reserve, as follows:
V

(t) = E
o
t
¸

n
t
e

s
t
r
o

l
ad
dN (s) −I (s) rds

+e

n
t
r
o
l
a
(n) I (n)
¸
.
V

(t) = E

t
¸

n
t
e

s
t
r

l
ad
dN (s) −I (s) rds

+e

n
t
r

l
a
(t) I (n)
¸
.
where the superscipts o and ∗ on E denote that N under these expectations
has the intensities µ
o
, the second order mortality rate, and µ

, the first order
mortality rate, respectively. We see that these quantities do not build on
the principles of no arbitrage and diversification. Indeed, we can trace both
discount factors and conditional expectations, but since the discount factor is
not based on the bank account interest rate and since the intensity of N in the
2.5 The liability and the payments 27
conditional expectation is not µ, these quantities can only be said to build on
suitable imitations of the principles.
2.5 The liability and the payments
In this section we introduce the idea of a payment process and discuss how
the market-based composition is built from payment processes of guaranteed
payments and payments which are not guaranteed.
The payments are the elements of the total payments in an insurance
contract. In our example, the payments are the elements of Equations (2.38)
and (2.39),
l
ad
dN (s) . l
a
(t) I (n) . −rI (s) ds and (l
a
(n) −l
a
(t)) I (n) .
The first task in connection with a precise description of the payments is
to identify the stochastic phenomena on which the claims depend. For certain
elementary insurances the time of death plays an important role. It turns out
to be beneficial to introduce the stochastic process Z based on the stochastic
time of death, given by Equation (2.33). The time of death is modeled by
a probability distribution of the counting process N, for example by the
introduction of a deterministic intensity µ.
To formalize claims we introduce a payment process B such that B(t)
represents the accumulated payments from the insurer to the insured over
|0. t]. This means that payments from the insured to the insurer appear in B
as negative payments. We specify the payments in continuous time, although
in practice the payments are discrete, for example monthly premiums. Fixing
the time horizon n for the insurance contract, we can describe the claims in
our example by collecting them in the payment process in the following form:
B(t) =

t
0
dB(s) . 0 ≤t ≤n. (2.42)
where
dB(t) =l
ad
dN (t) +l
a
(t) I (t) de (t. n) −rI (t) dt. (2.43)
Here, e (t. n) is an indicator process for t ≥ n. In Equation (2.42), the inte-
gration sums up the infinitesimal changes dB, and Equation (2.43) shows the
elements of these changes. We also speak of B as a payment stream, since it
describes payments floating between two parties.
The process Z serves to describe precisely a number of elementary claims
and payment processes in life and pension insurance, but there are many
28 Technical reserves and market values
0
active

(←)
1
disabled
2
dead
Figure 2.2. Disability model with recovery.
situations which are not described by such a process. One example is premium
waiver. Premium waiver and other types of disability assurances can be
described by extending the state space of Z with a third state, “disabled.”
One can, in general, let Z be a process moving around in a finite number of
states 1. The situation with a disability state is illustrated in Figure 2.2.
Corresponding to a general 1-state process Z, one can introduce a 1-
dimensional counting process where the ;th entry N
;
counts the number
of jumps into state ;. The disability model is a three-state model, i.e. 1 =3.
Models with more states are relevant for other types of insurances, for example
contracts on two lives where each member of a couple is covered against the
death of the other.
We mention this to give an idea of how the construction of payment
processes generalizes to various forms of disability payment processes etc.
All generalizations of liability formulas are left to the reader. An important
reference is Norberg (2000).
With the collection of claims in a payment process, we search for the
present value and the market value of a payment process. The principle of
no arbitrage determines the present value as a sum of present values of the
single elements of the payment process, and we get the present value at time
t of the payment process B over (t. n] as follows:

n
t
e

s
t
r
dB(s) .
Combining the principles of no arbitrage and diversification, we obtain the
market value at time t of a payment process as follows:
E
t
¸

n
t
e

s
t
r
dB(s)
¸
. (2.44)
2.5.1 The market-based liability revisited
In this section we identify the payment processes in the market-based com-
position of the liabilities.
2.5 The liability and the payments 29
In Equations (2.38)–(2.41) we wrote the entries in the market-based com-
position of the liability in terms of claims. We now collect these claims in
payment processes such that all entries can be written as market values of
payment processes. It is suitable to characterize the entries in the balance sheet
by different payment processes. This makes it easy to generalize the entries
to other insurances by simply specifying the generalized payment processes.
We also have in mind the situation where the condition X(n) =0 cannot be
fulfilled.
We introduce now the guaranteed payment process at time t, which we
denote by B(t. ·). Then for s ≥t,
dB(t. s) =−rI (s) ds +l
ad
dN (s) +l
a
(t) I (s) de (s. n) .
V
g
(t) =E
t
¸

n
t
e

s
t
r
dB(t. s)
¸
.
Correspondingly, we denote by B
b
(t. ·) the payment process not guaran-
teed at time t, and, for s ≥t,
dB
b
(t. s) =(l
a
(s) −l
a
(t)) I (s) de (s. n) .
V
b
(t) =E
t
¸

n
t
e

s
t
r
dB
b
(t. s)
¸
.
The difference V

(t) −V
g
(t) stems from the accumulated safety margins
in the guaranteed payments, denoted by C(t. ·), such that for s ≥t we have
dC(t. s) =I (s) c (t. s) ds.
V

(t) −V
g
(t) =E
t
¸

n
t
e

s
t
r
dC(t. s)
¸
. (2.45)
We remind ourselves that this constitutes the individual bonus potential if
U (t) ≥V

(t) ≥V (t).
We end this section by repeating, for the case U (t) ≥V

(t), the suggested
definition of the individual bonus potential in Equation (2.32), since Equation
30 Technical reserves and market values
(2.45) inspires us naturally to suggest two alternative definitions. Of course,
we have the following inequalities:
V
ib
(t) =

E
t
¸

n
t
e

s
t
r
dC(t. s)
¸
+
(2.46)
≤E
t
¸

n
t
e

s
t
r
dC(t. s)

+

(2.47)
≤E
t
¸

n
t
e

s
t
r
dC
+
(t. s)
¸
. (2.48)
Here, the quantities in Equations (2.47) and (2.48) comprise alternatives to
the definition given in Equation (2.46). The two alternatives coincide with the
first definition if c (t. s) (see Equation (2.29)) is either positive or negative
for all s. The definition of the individual bonus potential could be based on
a more precise description of how the safety margins are used for bonus. In
Section 2.6.3 we suggest a third alternative to Equation (2.46) which is not
based on the future safety margins.
The quantities in Equations (2.47) and (2.48) are closely connected to the
discussion about what an interest rate guarantee is worth. Whereas Equa-
tion (2.47) is connected to a terminal bonus, Equation (2.48) is connected to
a certain interpretation of the contribution principle. Values similar to Equa-
tion (2.48) are the basis for calculation of prices on interest rate guarantee
options.
2.6 The surrender option
2.6.1 Intensity-based surrender option valuation
In this section, we approach the value of the surrender option by the three-
state Markov model illustrated in Figure 2.3. Thus, we introduce a state of
0
alive and
insured

1
surrender
2
dead
Figure 2.3. Survival model with surrender option.
2.6 The surrender option 31
surrender, which at time t is reached by a surrender intensity ì and leads to
a payment of a surrender value G(t). Using a version of Thiele’s differential
equation for a three-state model, we can write the differential equation with
terminal condition for the total reserve V
sur
including the surrender option as
follows:
d
dt
V
sur
(t) = r (t) V
sur
(t) +r−µ(t)

l
ad
−V
sur
(t)

(2.49)
−ì (t) (G(t) −V
sur
(t)) .
V
sur
(n) = l
a
(n) .
We require an explicit expression for the value of the surrender option, i.e.
V
sur
−V. It can be verified that this additional reserve can be written in the
following form:
V
sur
(t) −V (t) =

n
t
e

s
t
r+µ+ì
ì (s) (G(s) −V (s)) ds. (2.50)
The calculation of V
sur
(t) −V (t) involves the future second order basis
through future values of G and V. Therefore, one is interested in a simplified
estimate of the value of the surrender option. We consider the case where the
surrender value equals the technical reserve, i.e. G(t) =V

(t), such that
V
sur
(t) −V (t) =

n
t
e

s
t
r+µ+ì
ì (s) (V

(s) −V (s)) ds. (2.51)
Firstly, we consider the case where V (t) - V

(t), i.e. X(t) - 0. We
assume that the future second order basis is settled such that o(s) ≤ c (s)
for s >t. One now obtains the following:
V
sur
(t) −V (t) =

n
t
e

s
t
r+µ+ì
ì (s) e

s
t
r+µ
(V

(t) −V (t)) ds
+

n
t
e

s
t
r+µ+ì
ì (s)

s
t
e

s
t
r+µ
(o(t) −c (t)) dt ds
≤(V

(t) −V (t)) ¡(t) . (2.52)
where
¡(t) =

n
t
e

s
t
ì
ì (s) ds.
Secondly, we consider the case where V (t) ≥ V

(t), i.e. X(t) ≥ 0. We
assume that the future second order basis is settled such that X(s) ≥ 0 for
32 Technical reserves and market values
s >t. We then obtain the following inequality:
V
sur
(t) −V (t) =

n
t
e

s
t
r+µ+ì
ì (s) (V

(s) −V (s)) ds
≤0. (2.53)
We conclude that, given the constraints on the future second order basis
leading to Equations (2.52) and (2.53), we have the following inequality:
V
sur
(t) −V (t) ≤max(¡(t) (G(t) −V (t)) . 0) .
2.6.2 Intervention-based surrender option valuation
In this section, we approach the value of the surrender option by considering
the idea that the insured surrenders “if it is worth it.” Hereby we take into
consideration the fundamental difference between transitions between the
states in the model illustrated in Figure 2.3. One can certainly criticize the
approach taken in this section. One problematic circumstance is that there
seems to be only weak historical evidence from situations where surrender
was optimal. Another circumstance is that, given a situation where surrender
is optimal, the insurer or the regulatory authorities will probably put up
a protection against systematic surrender. This can happen by either not
allowing surrender or by making surrender less attractive by introducing
costs. Will we in practice allow for a situation where the insured gains on
surrender? In this section the starting point is that the insured surrenders
immediately if it is advantageous to do so.
Optimal strategies of the insured provide a starting point for a detailed
study of the problem and a derivation of the corresponding deterministic
differential systems given by Steffensen (2002), which should be consulted
for a generalization of the payment processes, the ideas and the results in this
section.
Consider a general payment process B. By introducing the surrender option,
an arbitrage argument gives the market value including the surrender option
as follows:
V
sur
(t) = max
t≤t≤n
E
t
¸

t
t
e

s
t
r
dB(s) +e

t
t
r
I (t) G(t)
¸
. (2.54)
where we put G(n) =0. Here, we can think of B as the payment process for
the example given by Equation (2.43), and t is the time of surrender or the
terminal time, depending on what occurs first. Obviously the insured must
decide on surrender at time t based on the information available. One speaks
2.6 The surrender option 33
of t as a stopping time. We recall that I (t) indicates whether the insured is
alive at time t and that G(t) is the surrender value at time t.
In order to say something general about the quantity V
sur
(t), we consider
a process Y
sur
(t. u) defined by
Y
sur
(t. u) =

u
t
e

s
t
r
dB(s) +e

u
t
r
I (u) G(u) .
Thus, Y
sur
(t. u) is the present value of the payments to the insured including
the surrender value, given that the contract is surrendered at time u.
In the following, we use the notions of a super-martingale and a sub-
martingale, and we explain briefly these notions in the Markovian case.
A super-martingale describes the process that over time is expected to decrease
in value compared to where it is. Formally, the condition for Y to be a super-
martingale is that, for u ≥t,
E
t.x
|Y (u)] ≤x.
where the subscript t.x denotes that an expectation is conditional on]Y(t) =x].
Correspondingly, a sub-martingale describes a process expected to increase
in value. Formally, Y is a sub-martingale if, for u ≥t,
E
t.x
|Y (u)] ≥x.
Finally, a martingale is defined as a process which is both a sub-martingale
and a super-martingale, i.e. a process which is expected to maintain its value
over time. Formally, Y is a martingale if, for u ≥t,
E
t.x
|Y (u)] =x.
Proposition 2.1 (1) If
V(t) ≥G(t)
for all t, it is optimal never to surrender and
V
sur
(t) =V(t).
Intuition: if the total liability exceeds the surrender value, the company gains
by surrender the value V −G. Thus, it cannot be beneficial to surrender
and the company should set aside V, corresponding to the situation without
surrender option. If G≤ V

, we see that the classical solvency rule U ≥ V

exactly prevents the policy holder from making gains on surrender.
34 Technical reserves and market values
(2) If Y
sur
(t. u) is a sub-martingale, it is never optimal to surrender, and
V
sur
(t) =V(t).
Intuition: that Y
sur
(t. u) is a submartingale means that the present value
of payments to the policy holder is expected to increase as a function of
the time to surrender. Thus, the policy holder should at any point in time
postpone the surrender in order to increase the expected present value
of the contract.
(3) If Y
sur
(t. u) is a super-martingale, it is optimal to surrender immediately
and
V
sur
(t) =G(t).
Intuition: that Y
sur
(t. u) is a super-martingale means that the present
value of payments to the policy holder is expected to decrease as a
function of the time to surrender. Thus, the policy holder should surrender
immediately in order to obtain the highest possible expected present value
of the contract.
Given a value of the total payment process, for example given by Equa-
tion (2.54), the question is how to decompose this total reserve into a reserve
for the guaranteed payments, V
g.sur
, and a reserve for the payments which are
not guaranteed, V
b.sur
. One idea is to define V
g.sur
as follows:
V
g.sur
(t) = max
t≤t≤n
E
t
¸

t
t
e

s
t
r
dB(t. s) +e

t
t
I (t) G(t. t)
¸
. (2.55)
and then determine V
b.sur
residually as V
sur
−V
g.sur
. In Equation (2.55), G(t. t)
is the surrender value at time t given that no dividends are distributed over
(t. t]. We introduce the process Y
g.sur
(t. u) as follows:
Y
g.sur
(t. u) =

u
t
e

s
t
r
dB(t. s) +e

u
t
r
I (u) G(t. u) .
Thus, Y
g.sur
(t. u) is the present value of the guaranteed payments to the
insured including the surrender value given that the contract is surrendered at
time u.
Now we can state Proposition 2.1 with V
sur
(t), V (t) and Y
sur
(t. u) replaced
by V
g.sur
(t), V
g
(t), and Y
g.sur
(t. u). In order to apply this result, we need
to identify whether Y
g.sur
(t. u) is a super-martingale or a sub-martingale,
2.6 The surrender option 35
respectively. We use a result stating that if a process can be written as the
sum of a decreasing or increasing process and a martingale, then the process
itself is a super-martingale or a sub-martingale, respectively. This result is
applicable once we have established the following lemma.
Lemma 2.2 Assume that G(t. u) = V

(t. u). Then it is possible to write
Y
g.sur
(t. u) as follows:
Y
g.sur
(t. u) =V

(t) −

u
t
e

s
t
r
I (s) c (t. s) ds +M
sur
(t. u) .
where M
sur
(t. s) is a martingale.
We can now conclude the following for the case G(t. u) = V

(t. u) as
follows.
Corollary 2.3 (1) If c (t. s) ≥0, we have that
V
g.sur
(t) =V

(t).
Intuition: the policy holder loses the future positive safety margins by
keeping the contract. This can be avoided by immediate surrender such
that the value of the guaranteed payments simply becomes the surrender
value.
(2) If c (t. s) ≤0, we have that
V
g.sur
(t) =V
g
(t).
Intuition: the policy holder has the negative safety margins covered by
keeping the insurance contract since we disregard all bonus payments.
The policy holder gains maximally in expectation by keeping the policy
until termination, such that the value of the guaranteed payments simply
becomes the corresponding value without the surrender option.
In Proposition 2.1 and Corollary 2.3 we put up specific conditions under
which it is relatively easy to calculate the relevant maximum. Nevertheless,
it is possible to derive a deterministic differential system, comparable with
Thiele’s differential equation, for calculation of the general market values
V
sur
and V
g.sur
. Even though this goes beyond the scope of this chapter, we
36 Technical reserves and market values
give, out of interest, the result for V
g.sur
. We have that V
g.sur
(t) =V
g.sur
(t. t),
where
o
ou
V
g.sur
(t. u) ≤r+r (u) V
g.sur
(t. u) −µ(u)

l
ad
−V
g.sur
(t. u)

.
V
g.sur
(t. t) ≥G(t) .
0 =


o
ou
V
g.sur
(t. u) +r+r (u) V
g.sur
(t. u)
−µ(u)

l
ad
−V
g.sur
(t. u)

×(G(t. u) −V
g.sur
(t. u)) .
V
g.sur
(t. n) =l
a
(t) .
The reader should recognize several elements in this system. The first inequal-
ity, together with the terminal condition, only differs from Thiele’s differen-
tial equation by containing an inequality instead of an equality. The second
inequality simply states that the sum at risk connected to the transition to
the state surrender is always negative. The equality in the third line is the
formalization of the following statement: at any point in the state space, one
of the two inequalities is an equality.
2.6.3 The market-based liability revisited
Corollary 2.3 is closely connected to the market-based composition of the
liabilities. We repeat the definition of the individual bonus potential given
in Equation (2.31) for the case V ≥ V

, since the studies in Section 2.6.2
lead us, quite naturally, to an alternative definition. We have the following
inequality:
V
ib
=max(V

. V
g
) −V
g
≤V
g.sur
−V
g
. (2.56)
with G(t. u) =V

(t. u). Corollary 2.3 shows us that the alternative definition
in Equation (2.56) coincides with the definition in Equation (2.31) if c (s. t)
is either positive or negative for all s. In Equations (2.47) and (2.48) we gave
alternatives to the definition in Equation (2.31) in connection with a discussion
of how the safety margins are redistributed as dividends. In Equation (2.56)
we suggest an alternative which does not contain this subjectivity of the
dividend distribution.
2.7 The free policy option 37
Actually, it is the inequality in Equation (2.56) which motivates us to sug-
gest the alternative definition, V
g.sur
−V
g
, for the individual bonus potential.
The inequality in Equation (2.56) follows from
max(V

. V
g
) ≤V
g.sur
.
which is true since the left hand side corresponds to Equation (2.55), where
the maximum is taken over only two possible points in time, t and n. On
the right hand side, the maximum is taken over all points in time between
t and n.
2.7 The free policy option
2.7.1 A simple free policy option value
The starting point for the value of the free policy option is a model with three
states as illustrated in Figure 2.4. However, the idea is not to introduce a free
policy intensity in the same way as we introduced a surrender intensity in
Section 2.6.1. We recall that
V

(t) −V
g
(t) =

n
t
e

s
t
r+µ
c (t. s) ds.
A part of the safety margin in the guaranteed payments can be said to stem
from future premiums. One can argue that if we take into account the future
premiums we should also, given the free policy option, put up a reserve for
the bonus based on these premiums.
The liability concerning bonus stemming from the future premiums is
given by the bonus potential on a contract with premium r issued at time t
with a deposit equal to zero. We now calculate this bonus potential and start
0
alive and
with premium

1
free policy
2
dead
Figure 2.4. Survival model with free policy option.
38 Technical reserves and market values
out by finding the benefits on the issued contract. We introduce the following
notation:
V
∗+
(t) =l
ad
A
1∗
x+tn−t
+l
a
(t)
n−t
E

x+t
. (2.57)
V
g+
(t) =l
ad
A
1
x+tn−t
+l
a
(t)
n−t
E
x+t
. (2.58)
i.e. the superscript + denotes here that only the benefits are included. Then
we can, from Equations (2.9) and (2.22), write the life annuities o

x+tn−t
and
o
x+tn−t
in the following way:
o

x+tn−t
=
V
∗+
(t) −V

(t)
r
. (2.59)
o
x+tn−t
=
V
g+
(t) −V
g
(t)
r
. (2.60)
The number of units of guaranteed benefits, 8(t), that can be bought from the
future premiums at time t are defined simply by the equivalence principle:
8(t)

l
ad
A
1∗
x+tn−t
+l
a
(t)
n−t
E

x+t

=ro

x+tn−t
.
which, by Equations (2.57) and (2.59), yields
8(t) =
V
∗+
(t) −V

(t)
V
∗+
(t)
. (2.61)
In Section 2.3.2 we noted that the bonus potential on an insurance contract
at the time of issuance simply equals the negative market value of the guar-
anteed payments stipulated in the contract. We see that the bonus potential
on the new insurance is given by
−8(t)

l
ad
A
1
x+tn−t
+l
a
(t)
n−t
E
x+t

+ro
x+tn−t
.
which by Equations (2.58), (2.60) and (2.61) equals
V

(t) V
g+
(t)
V
∗+
(t)
−V
g
(t) .
This quantity is now our candidate for the value of the free policy option. This
value must be added to the market value of the guaranteed original payments
in order to obtain the value of the guaranteed payments including the free
policy option. We reach the following:

V

(t) V
g+
(t)
V
∗+
(t)
−V
g
(t)

+V
g
(t) =
V

(t) V
g+
(t)
V
∗+
(t)
≡V
f
(t) .
2.7 The free policy option 39
Note that V
f
is also the market value of the guaranteed free policy benefits
since the guaranteed benefits upon conversion into a free policy are exactly
reduced by the factor V

(t) ¡V
∗+
(t).
It turns out to be informative to decompose the individual bonus potential
into the bonus potential on premiums V
bp
and the bonus potential on the free
policy V
bf
. Firstly, we consider the situation
V ≥V

≥V
f
≥V
g
.
Then V
ib
=V

−V
g
decomposes into
V
ib
=V
bf
+V
bp
=

V

−V
f

+

V
f
−V
g

.
The market value of guaranteed payments on a contract issued at time
t, V
bp
(t), can also be expressed as the market value of the safety margins
in these payments. A situation can arise where these safety margins are
negative. Since negative safety margins cannot be distributed as bonus, the
bonus potential on premiums is in this case set to zero. For the case V ≥V

we obtain an expression for V
bp
which is similar to Equation (2.31):
V
bp
=max

V
f
. V
g

−V
g
. (2.62)
If, at one time, we want to take into consideration all possible relations
between V, V

, V
f
and V
g
, the bonus potential on premiums and the bonus
potential on the free policy can be formalized as follows:
V
bf
=

V −V
g

V
f
−V
g

+

+

V −V
g
−(V

−V
g
)
+

+
.
V
bp
=V −V
g

V −V
g

V
f
−V
g

+

+
.
2.7.2 Intervention-based free policy option valuation
In this section the free policy option is dealt with in the same way as the
surrender option was dealt with in Section 2.6.2. Even though the intuitive
interpretations on this background could be left to the reader, we keep them
here and follow Section 2.6.2 closely.
We now take the logical starting point that the insured converts into free
policy “if it is worth it.” Hereby we take into consideration the fundamental
difference between transitions between the states in the model illustrated in
Figure 2.4. One can certainly criticize the approach taken in this section.
One problem is that we have no historical evidence from a situation where
40 Technical reserves and market values
conversion into a free policy was optimal. Another is that, given a situation
where conversion into free policy is optimal, the insurer or the regulatory
authorities will probably put up a protection against systematic conversion.
This can happen by making conversion less attractive by introducing costs.
In opposition to the surrender option, however, one cannot stop conversions
into free policy. This would involve difficult considerations about credit risk.
Thus, will one in practice allow for a situation where the insured gains on
conversion? In this section the starting point is that the insured converts
immediately if it is advantageous to do so.
Optimal strategies of the insured provide a starting point for a detailed
study of the problem and a derivation of the corresponding deterministic
differential systems given by Steffensen (2002), which should be consulted
for a generalization of the payment processes, the ideas and the results in this
section.
Consider a general payment process B. By introducing the free policy
option, an arbitrage argument gives the market value including the free policy
option as follows:
V
free
(t) = max
t≤t≤n
E
t
¸

t
t
e

s
t
r
dB(s) +e

t
t
r
I (t)

V
f
(t) +X
f
(t)

¸
. (2.63)
where V
free
corresponds to V
sur
with the surrender value G(t) replaced by
the reserve V
f
(t) +X
f
(t). We think of B as the payment process for the
example given by Equation (2.43), and t is the time of conversion or the
terminal time depending on what occurs first. Obviously the insured must
decide on conversion at time t based on the information existing at that time.
Thereby, t is a stopping time. The quantity X
f
(t) is the undistributed reserve
upon conversion, i.e. if the full undistributed reserve is carried over upon
conversion, X
f
=X.
In order to say something general about the quantity V
free
(t), we consider
a process Y
free
(t. u) as a function of u, which we define by
Y
free
(t. u) =

u
t
e

s
t
r
dB(s) +e

u
t
r
I (u)

V
f
(u) +X
f
(u)

.
where
V
f
(t) =
V

(t) V
g+
(t)
V
∗+
(t)
.
Thus, Y
free
(t. u) is the present value of the payments to the insured, including
the liability to cover free policy benefits and benefits which are not guaranteed
from the time of conversion, given that conversion happens at time u.
2.7 The free policy option 41
Proposition 2.4 (1) If
V(t) ≥V
f
(t) +X
f
(t)
for all t, it is optimal never to convert and
V
free
(t) =V(t).
Intuition: if the total liability exceeds the market value of payments at con-
version, the company gains by conversion the value V −

V
f
(t) +X
f
(t)

.
Thus, it cannot be beneficial to convert and the company should set
aside V, corresponding to the situation without free policy option. If
V
f
+X
f
≤ V

, we see that the classical solvency rule U ≥ V

exactly
prevents the policy holder from making gains on conversion.
(2) If Y
free
(t. u) is a sub-martingale, it is never optimal to convert and
V
free
(t) =V(t).
Intuition: that Y
free
(t. u) is a sub-martingale means that the present value
of payments to the policy holder is expected to increase as a function
of the time to conversion. Thus, the policy holder should at any point in
time postpone the conversion in order to increase the expected present
value of the contract.
(3) If Y
free
(t. u) is a super-martingale, it is optimal to convert immediately
and
V
free
(t) =V
f
(t) +X
f
(t).
Intuition: that Y
free
(t. u) is a super-martingale means that the present
value of payments to the policy holder is expected to decrease as a
function of the time to conversion. Thus, the policy holder should convert
immediately in order to obtain the highest possible expected present value
of the contract.
Given a value of the total payment process, for example given by Equation
(2.63), the question is how to decompose this total reserve into a reserve for
the guaranteed payments, V
g.free
, and a reserve for the payments which are
not guaranteed, V
b.free
. We suggest the definition
V
g.free
(t) = max
t≤t≤n
E
t
¸

t
t
e

s
t
r
dB(t. s) +e

t
t
I (t) V
f
(t. t)
¸
. (2.64)
42 Technical reserves and market values
where
V
f
(t. t) =
V

(t. t) V
g+
(t. t)
V
∗+
(t. t)
.
Then one could choose to define V
b.free
residually by V
b.free
= V
free
−V
g.free
.
We introduce the process Y
g.free
(t. u) given by
Y
g.free
(t. u) =

u
t
e

s
t
r
dB(t. s) +e

u
t
r
I (u) V
f
(t. u) .
Thus, Y
g.free
(t. u) is the present value of the guaranteed payments to the
insured, including the liability for future free policy benefits, given that the
contract is converted at time u.
Nowwe can state Proposition 2.4 with V
free
(t), V
f
(t)+X
f
(t) and Y
free
(t. u)
replaced by V
g.free
(t), V
g
(t) and Y
g.free
(t. u). In order to apply this result, we
need to identify when Y
g.free
(t. u) is a super-martingale or a sub-martingale,
respectively. We use a result stating that if a process can be written as the
sum of a decreasing or increasing process and a martingale, then the process
itself is a super-martingale or a sub-martingale, respectively. This result is
applicable once we have established the following lemma.
Lemma 2.5 It is possible to write Y
free
(t. u) as follows:
Y
free
(t. u) =V
f
(t) −

u
t
e

s
t
r
I(s)c
free
(t. s) ds +M
free
(t. u) .
where M
free
(t. u) is a martingale and c
free
(t. u) is a rather involved function of
V

(t), V
∗+
(t), V
g
(t), V
g+
(t) and the first order and real bases. See Steffensen
(2002) for details of the function c
free
(t. u).
We can now conclude the following.
Corollary 2.6 (1) If c
free
(t. u) ≥0, we have that
V
g.free
(t) =V
f
(t).
Intuition: the policy holder loses the future positive safety margins on
the future premiums by continuing to pay the premiums on the contract
since we disregard all future bonus payments. This can be avoided by
immediate conversion such that the value of the guaranteed payments
simply becomes the market value of the free policy benefits.
2.7 The free policy option 43
(2) If c
free
(t. u) ≤0, we have that
V
g.free
(t) =V
g
(t).
Intuition: the policy holder has the negative safety margins on future
premiums covered by continuing to pay the premiums on the insurance
contract since we disregard all bonus payments. The policy holder gains
maximally in expectation by continuing to pay the premium on the policy
until termination, such that the value of the guaranteed payments simply
becomes the corresponding value without the free policy option.
In Proposition 2.4 and Corollary 2.6 we put up specific conditions under
which it is relatively easy to calculate the relevant maximum. Nevertheless,
it is possible to derive a deterministic differential system, comparable with
Thiele’s differential equation, for calculation of the general market values V
free
and V
g.free
. Even though this goes beyond the scope of this chapter, we give, out
of interest, the result for V
g.free
. We have that V
g.free
(t) =V
g.free
(t. t), where
o
ou
V
g.free
(t. u) ≤r+r (u) V
g.free
(t. u) −µ(u)

l
ad
−V
g.free
(t. u)

.
V
g.free
(t. t) ≥V
f
(t) .
0 =


o
ou
V
g.free
(t. u) +r+r (u) V
g.free
(t. u)
−µ(u)

l
ad
−V
g.free
(t. u)

×

V
f
(t. u) −V
g.free
(t. u)

.
V
g.free
(t. n) =l
a
(t) .
The reader should recognize several elements in this system. The first inequal-
ity together with the terminal condition, only differs from Thiele’s differen-
tial equation by containing an inequality instead of an equality. The second
inequality simply states that the sum at risk connected to the transition to
the free policy state is always negative. The equality in the third line is the
formalization of the following statement: at any point in the state space, one
of the two inequalities is an equality.
2.7.3 The market-based liability revisited
Corollary 2.6 is closely connected to the market-based composition of the
liabilities. We repeat the definition of the individual bonus potential on pre-
miums given in Equation (2.31) for the case V ≥ V

, since the studies in
44 Technical reserves and market values
Section 2.7.2 lead us, quite naturally, to suggest an alternative definition. We
have the following inequality:
V
bp
= max

V
f
. V
g

−V
g
≤ V
g.free
−V
g
. (2.65)
Corollary 2.6 shows us that the alternative definition in Equation (2.65)
coincides with the definition in Equation (2.62) if c
free
(t. s) is either positive
or negative for all s.
Actually, it is the inequality in Equation (2.65) which motivates us to
suggest the alternative definition, V
g.free
−V
g
, for the bonus potential on
premiums. The inequality in Equation (2.65) follows from
max

V
f
. V
g

≤V
g.free
.
which is true since the left hand side corresponds to Equation (2.64), where
the maximum is taken over only two possible points in time, t and n. On
the right hand side, the maximum is taken over all points in time between t
and n.
3
Interest rate theory in insurance
3.1 Introduction
This chapter provides a brief introduction to some basic concepts from interest
rate theory and financial mathematics and applies these theories for the cal-
culation of market values of life insurance liabilities. There exists a huge
amount of literature on financial mathematics and interest rate theory, and we
shall not mention all work of importance within this area. Some basic intro-
ductions are Baxter and Rennie (1996) and Hull (2005). Readers interested
in more mathematical aspects of these theories are referred to Lamberton and
Lapeyre (1996) and Nielsen (1999). Finally we mention Björk (1997, 2004)
and Cairns (2004).
The present chapter is organized as follows. Section 3.2 demonstrates how
the traditional actuarial principle of equivalence can be modified in order to
deal with situations with random changes in the future interest rate, i.e. to
the case of stochastic interest rates. This argument, which involves hedging
via so-called zero coupon bonds, leads to new insights into the problem of
determining the market value for the guaranteed payments on a life insurance
contract. Section 3.3 gives a more systematic treatment of topics such as zero
coupon bonds, the term structure of interest rates and forward rates. In addi-
tion, this section demonstrates how versions of Thiele’s differential equation
can be derived for the market value of the guaranteed payments. In these equa-
tions, forward rates now appear instead of the interest rates. In Section 3.5,
we mention more general bonds and relations to zero coupon bonds, and in
Section 3.6 we briefly turn to the estimation of forward rates. Section 3.7
presents an introduction to arbitrage pricing theory in models, where trading
is possible at fixed discrete time points. This discussion leads to formulas
for market values involving so-called martingale measures. In addition, the
analysis provides a basis for an integrated description of the liabilities and
45
46 Interest rate theory in insurance
assets of the insurance company which can be used for assessment of the com-
pany’s total financial situation under various economic scenarios. Essential
elements are stochastic models for the future interest rate, bond prices and
stock prices. The importance of this analysis is increased by the introduction
of market-based accounting in life insurance, since in this case the market
value of the liabilities is also affected by the financial situation.
3.2 Valuation by diversification revisited
How are life insurance contracts traditionally valuated? What is the reasoning
behind these methods? What is diversification? To answer these questions, let
us first recall the traditional actuarial law of large number considerations and
see how these can be modified for the most simple life insurance contracts
in the presence of financial uncertainty. In particular, we consider situations
with stochastic interest.
3.2.1 The law of large numbers
A first version of the law of large numbers was formulated by the Swiss math-
ematician Jakob Bernoulli. The law of large numbers states that the average
from some experiment (for example, the average number of heads from toss-
ing the same coin m times) will converge to some number as the number of
experiments increases. In the coin-tossing example, we could let X
i
be the
outcome from the ith toss, by taking X
i
= 1 if the coin shows a “head” and
X
i
= 0 if it shows a “tail.” We can write this in a more compact form using
the indicator function X
i
= 1
A
i , where A
i
is the event that the ith toss leads to
a “head,” i.e. A
i
=]ith coin head]. Then the lawof large numbers states that
1
m
m
¸
i=1
1
A
i →E|1
A
1 ]. (3.1)
as m increases to infinity. Since the expected value of the indicator function
1
A
1 is equal to the probability of the event A
1
, we see that the average of
heads converges to the probability of observing a head, which for most coins
should be
1
2
.
3.2.2 A portfolio of insured
Consider now a portfolio of /
x
, say, identical n-year pure endowment contracts
with sum insured 1. For t ≥ 0 we denote by /
x+t
the expected number of
survivors at age x+t. We impose here the following standard assumptions.
3.2 Valuation by diversification revisited 47
• The /
x
policy holders are all of age x at time 0 with remaining lifetimes
given by T
1
. . . . . T
/
x
with the same survival probability,
t
¡
x
=
/
x+t
/
x
=exp

t
0
µ(x+u) du

. (3.2)
where the mortality intensity µ is a deterministic function, and where
we have used standard actuarial notation. For example, µ could be a
Gompertz–Makeham intensity of the form
µ(x+t) =o+8c
x+t
.
• The contracts are paid by a single premium r(0) at time 0.
Of course, we cannot predict exactly the number of policy holders, /
x
, who
actually survive until time n, at which time each survivor receives one unit.
However, we know that if the portfolio is large, the actual number of survivors
at time n is in some sense close to the expected number /
x+n
. Mathematically,
this is a consequence of the law of large numbers. To see this, introduce the
indicator 1
]T
i
>n]
that the ith policy holder survives. The ratio between the
actual (unknown) number of survivors at time n and the number of policy
holders /
x
entering the contract at time 0 can now be written as follows:
1
/
x
/
x
¸
i=1
1
]T
i
>n]
. (3.3)
The situation is now almost identical to the coin-tossing example, and, by the
law of large numbers, Equation (3.3) converges to
E|1
]T
1
>n]
] =P(T
1
>n) =
n
¡
x
=
/
x+n
/
x
as the number of policy holders /
x
increases to infinity, provided that the
lifetimes of the policy holders are independent. This shows that the actual
number of survivors is indeed close to the number predicted by the decrement
series /
x+n
if /
x
is “big,” i.e.
/
x
¸
i=1
1
]T
i
>n]
≈/
x n
¡
x
=/
x+n
.
In a sufficiently large insurance portfolio, the actual number of survivors at a
given fixed time is hence relatively close to the expected number of survivors.
Similar arguments can be applied in situations where several time points or
even payment processes are considered.
We emphasize that the assumed independence is crucial in order to obtain
this convergence. In more realistic models, where the future mortality inten-
sity µ is unknown (stochastic), the law of large numbers does not imply that
48 Interest rate theory in insurance
the quantity (3.3) converges to a constant. Here, one distinguishes between
unsystematic and systematic mortality risk. The systematic risk is associated
with the consequences of random changes in the underlying mortality inten-
sity, and the unsystematic risk is the risk associated with the insured lifetimes
given the underlying mortality intensity. It is the unsystematic risk, which is
diversifiable and can be eliminated by increasing the size of the portfolio,
whereas the systematic is undiversifiable and remains with the insurer. This
problem is addressed further in Section 5.5.
3.2.3 Interest, accumulation and discount factors
What is the relation between yearly interest rate and the force of interest?
How do we handle interest rates that are not constant during the term of the
contracts? In this section, we recall these and other basic concepts.
In the traditional actuarial literature, see for example Gerber (1997), it is
typically assumed (implicitly) that the insurance company invests capital in an
account with yearly interest rate i, and that this rate is constant during the term
of the insurance contract. Introducing r =log(1+i), we can alternatively write
the one-year accumulation factor 1 +i as e
r
. Thus, the t-year accumulation
factor is determined as follows:
(1+i)
t
=e
rt
=S(t). (3.4)
which clearly satisfies the differential equation
d
dt
S(t) =rS(t). (3.5)
with the initial condition S(0) = 1. The quantity r is known as the force
of interest, and it can be interpreted as the interest per time unit per unit
deposited on the account. The t-year discount factor is given by
v
t
=(1+i)
−t
=e
−rt
=S(t)
−1
. (3.6)
where v =e
−r
. How can this be generalized to situations where i is no longer
constant? One possibility is to let i(s) represent the annual interest rate for
year s. The corresponding force of interest r(s) for year s is then determined
from the equation (1 +i(s)) = e
r(s)
, and the t-year accumulation factor is
given by
S(t) =(1+i(1)) · · · (1+i(t)) =exp

t
¸
s=1
r(s)

. (3.7)
3.2 Valuation by diversification revisited 49
If we consider instead the case where the interest rate can change more often
than once every year, then the force of interest is the most natural starting
point, since it describes the interest per time unit. Letting r(u) be the force of
interest at any time u, we can define a new accumulation factor by changing
Equation (3.5) such that r depends on time, that is
d
dt
S(t) =r(t)S(t). (3.8)
It does not look like a big difference compared to Equation (3.5), but if we
solve Equation (3.8) with the initial condition S(0) = 1, the accumulation
factor becomes
S(t) =exp

t
0
r(u) du

. (3.9)
and the corresponding discount factor is given by
S(t)
−1
=exp

t
0
r(u) du

. (3.10)
Note that the quantity S(t) represents the value at time t of one unit invested at
time 0 in the account which bears continuously added interest r(u). In financial
mathematics, one refers to this account as a savings account.
We use the accumulation and discount factors from Equations (3.9) and
(3.10) in the following. One advantage compared to working with a constant
interest rate is that we can now allow interest rates to vary over time by spec-
ifying the force of interest as a function of time. In addition, this framework
can be used when building models which allow for random changes in the
future interest rates. This extension is the basis for more realistic studies of
the impact of changes in the interest rates on the balance sheet of an insurance
company, an area which is essential within asset-liability management.
3.2.4 The insurer’s loss and the principle of equivalence
Now consider a portfolio of /
x
pure endowments of one unit expiring at time
n paid by a single premium r(0) at time 0. Assume moreover that the number
of survivors follows the decrement series, i.e. it is deterministic and equal to
/
x+n
. With this contract, each of the survivors at time n receive one unit; the
present value at time 0 of this benefit is given by S(n)
−1
/
x+n
. Since the premiums
/
x
r(0) are payable at time 0, no discounting is needed, and the present value
of the premiums is simply equal to the payment /
x
r(0). Thus, the present value
of the insurer’s loss associated with the portfolio can be defined as follows:
L =/
x+n
S(n)
−1
−/
x
r(0). (3.11)
50 Interest rate theory in insurance
The premium r(0) is now said to be fair, if L =0. If S(n) is deterministic (or
known at time 0), the fair premium is the well known equivalence premium:
using the fact that S(t) = exp(

t
0
r(u) du), we get from Equation (3.11) the
following well known result:
r(0) =
/
x+n
/
x
exp

n
0
r(u) du

=
n
¡
x
exp

n
0
r(u) du

. (3.12)
which is simply the expected present value of the benefit. However, it is
important to realize that this argument works only in the situation where
the discount factor S(n)
−1
= exp(−

n
0
r(u) du) is deterministic. In the case
where the future interest rates are unknown at time 0, we cannot charge
this premium, since we do not know S(n) at the time of selling of the
contract!
Thus, we see from the simple considerations above that mortality risk can
essentially be eliminated by increasing the size of the portfolio if the insured
lives are independent, i.e. if there is no systematic insurance risk within
the model. In this situation, we also say that mortality risk is diversifiable.
However, we see from (3.12) that we did not get rid of the factor S(n)
−1
in
the case where S(n) is random. We can interpret this by saying that we have
eliminated the mortality risk, whereas the financial risk related to the future
development of the interest rate could not be eliminated by increasing the size
of the portfolio. This is not so surprising, but it is nevertheless worth pointing
out, since it raises the question of how we can control or eliminate this risk.
The answer is straightforward in the idealized world of our simple example:
by trading with financial contracts called zero coupon bonds. (Sections 3.3
and 3.5 below are devoted to a more systematic treatment of this topic.) A zero
coupon is a contract which pays its holder one unit at a fixed time n (also
referred to as an n-bond); the price at time 0 ≤ t ≤ n is denoted by P(t. n).
Clearly, we must have that P(n. n) =1. Before analyzing the company’s loss
in this setting, we consider the special case where all future zero coupon
bond prices are known at time 0, i.e. the situation with deterministic bond
prices.
3.2.5 Deterministic bond prices
Let us consider a completely deterministic world, i.e. there is no randomness,
where tomorrow’s prices for zero coupon bonds are known today. In this
setting, we can derive the structure of zero coupon bond prices and give
another introduction to the concept of arbitrage.
3.2 Valuation by diversification revisited 51
Fix some times t ≤n, where t is today and n is the payment time. If bond
prices are deterministic, the price P(t. n) at time t of an n-bond is known
already at time t for any t ∈ |t. n]. In particular, this implies that
P(t. n) =P(t. t) P(t. n). (3.13)
which says that the value at time t of an n-bond is equal to the value at
time t of a t-bond multiplied by the value at time t of an n-bond. Thus,
one can in this case think of P(t. t) and P(t. n) as discount factors for |t. t]
and |t. n], respectively. To show that Equation (3.13) is satisfied, recall that
P(t. n) is the price at time t of one unit at time n. Under the assumption of
deterministic zero coupon bonds, one can alternatively obtain one unit at time
n by investing in P(t. n) t-bonds at the price P(t. t) P(t. n). At time t, this
leads to the payment P(t. n), which can be used to buy an n-bond, and this
ensures the payment of one unit at time n. This shows that P(t. t) and P(t. n)
can be interpreted as discount factors. However, it is essential to realize that
Equation (3.13) is not satisfied in the more realistic situation where P(t. n)
is not known at time t for t > t. The reason is again that P(t. n) and P(t. t)
are known at time t, whereas P(t. n) is not known at t, when t >t.
Assuming in addition that the zero coupon prices are sufficiently nice
(smooth) functions, one can actually prove that Equation (3.13) implies the
existence of a function r

such that
P(t. n) =exp

n
t
r

(u) du

. (3.14)
An alternative way of addressing this issue is as follows. Assume that
we can buy zero coupon bonds with maturity n at the price given by Equa-
tion (3.14) and that we can invest in the savings account with deterministic
interest r(u) as explained in Section 3.2.3. We have thus constructed a market
with two investment possibilities (two assets). It can now be shown that if
we insist on prices which do not allow for the possibility of risk-free gains,
then there is only one possible price for the zero coupon bond. This type of
argument is also called an arbitrage argument. More precisely, we show that
the only reasonable price for the bond is given by
P(t. n) =exp

n
t
r(u) du

=
S(t)
S(n)
. (3.15)
so that r

(u), which appears in Equation (3.14), must be identical to the
interest rate r(u) from the savings account. For reasons of completeness, we
give this argument here.
52 Interest rate theory in insurance
Assume that the price of an n-bond at time t differs from that given in
Equation (3.15) and is given by
P(t. n) =(1+a) exp

n
t
r(u) du

=(1+a)
S(t)
S(n)
.
for some a =0. It is now possible to construct a risk-free gain by using the
following strategy (if a >0).
• Sell one bond at the price P(t. n) at time t. This leads to the liability 1 at
time n.
• Invest the amount P(t. n) in the savings account at time t. At time n, the
deposit on the savings account is then P(t. n)S(n)¡S(t) =(1+a).
• At time n, withdraw the amount (1+a) from the savings account and pay
one unit to the buyer of the bond.
This strategy leads to a gain of a. If a - 0, a gain of −a can be obtained
by borrowing money from the bank and buying a bond. The main idea of
arbitrage-free pricing is that such possibilities cannot exist in the market,
since all investors would want to sell the bond if a > 0, and there would be
no buyers. Consequently, prices would adapt such that it is no longer possible
to generate risk-free gains.
We point out that this argument cannot be applied in the situation where
r(u) is not deterministic. This can be seen directly by considering Equa-
tion (3.15), which involves the future (unknown) interest rate. One alternative
idea could be to replace S(t)¡S(n) by its expected value. However, as we see
below, there is no reasonable argument which supports this idea. Finally, we
note that in the case where r(u) is constant and equal to r, Equation (3.15)
simplifies to
P(t. n) =exp(−r(n−t)).
which is identical to the classical discount factor given in Equation (3.6).
3.2.6 Hedging with zero coupon bonds
In this section, we modify the principle of equivalence to the situation where
the insurer can trade zero coupon bonds. This gives a way of hedging, i.e.
controlling or eliminating, the risk associated with the development of the
interest rate. We consider in this section only the guaranteed payments as
defined in Chapter 2; the treatment of payments related to bonus is postponed
until Chapter 4.
3.2 Valuation by diversification revisited 53
Hedging a portfolio of pure endowments
Assume that the insurer at time 0 invests in /
x
« units of an n-bond. The
present value of the insurer’s loss from the portfolio of /
x
pure endowments
is then given by
˜
L =(/
x+n
S(n)
−1
−/
x
r(0)) +/
x
«(P(0. n) −1· S(n)
−1
). (3.16)
where we have assumed that the number of survivors follows the decre-
ment series. In addition, we have discounted payments by using the true
interest rate (the market interest rate). The first term in Equation (3.16) cor-
responds to the present value of the company’s loss without investments in
n-bonds (but investment in the savings account), and the second term repre-
sents the present value of the loss from buying at time 0 exactly /
x
«n-bonds
at the price P(0. n). This term is the difference between the price P(0. n) of
the bond at time 0 and the discounted amount received by the company at
time n. By rearranging terms in Equation (3.16), we see that
˜
L =(/
x+n
−/
x
«)S(n)
−1
+/
x
(«P(0. n) −r(0)).
Here, the first term is equal to zero exactly if « = /
x+n
¡/
x
=
n
¡
x
, and the
second term is zero if, in addition,
r(0) =
n
¡
x
P(0. n). (3.17)
Thus we have obtained that
˜
L =0, as in the classical situation with determin-
istic interest. The fair premium given by Equation (3.17) has a very natural
form: it is the price at time 0 of a zero coupon bond with maturity n multiplied
by the probability of survival to n. In the portfolio with /
x
policy holders, the
insurer should purchase /
x n
¡
x
= /
x+n
bonds, which is exactly the expected
number of survivors.
Note that the arguments leading to Equation (3.17) determine the fair
premium (market price) for the guaranteed payments as well as an investment
strategy: that the insurer should invest the entire premium (the market value)
in n-bonds. In this way, the insurer is able to replicate the liabilities, since
the value at a future time t of the bonds purchased at time 0 is given by
/
x n
¡
x
P(t. n) =/
x+n
P(t. n).
The argument used for determining the price at time 0 of the guaranteed
part of the pure endowment can now be repeated at time t for each of the
/
x+t
remaining policy holders. Hence, the total market value at time t of the
54 Interest rate theory in insurance
guaranteed payments associated with these /
x+t
pure endowment contracts is
given by
/
x+t n−t
¡
x+t
P(t. n) =/
x+n
P(t. n).
which shows that the market value of the liabilities is exactly equal to the
value of the investments (the assets) at any time t for any future development
of the value of the zero coupon bonds. Any increase or decrease of the
bond price leads to exactly the same changes in the value of the assets and
liabilities.
Note that the market value of the guaranteed payments does not depend on
the company’s choice of investment strategy, even though the market value
is derived by means of a hedging argument. If it were possible to purchase
and sell such contracts at prices which deviated from the market value, it
would in fact be possible to generate risk-free gains (arbitrage) by investing
in zero coupon bonds. For a more detailed treatment of these aspects, see
Møller (2000) and Steffensen (2001).
Hedging a portfolio of term insurances
For completeness, we indicate how this hedging argument can be applied for
the pricing and hedging of the guaranteed part of a term insurance. Consider
a portfolio of /
x
term insurances. The number of deaths in year t predicted at
time 0 by the decrement series is given by
J
x+t
=/
x+t
−/
x+t+1
.
where we have used traditional actuarial notation. Consider for simplicity the
case where the sum insured (one unit) is payable at the end of the year, i.e. at
times t =1. 2. . . . . n. In this case, the argument used for the pure endowment
has to be modified slightly, so that the company now invests in zero coupon
bonds with expiration times t = 1. 2. . . . . n. Assume more precisely that
the company at time 0 invests in /
x
«(t) t-bonds at the price P(0. t). The
company’s loss at time 0 then becomes
˜
L =

n
¸
t=1
J
x+t−1
S(t)
−1
−/
x
r(0)

+/
x
n
¸
t=1
«(t)

P(0. t) −1· S(t)
−1

. (3.18)
which can be interpreted in the same way as for the pure endowment. This
expression can be rewritten as follows:
˜
L =
n
¸
t=1
(J
x+t−1
−/
x
«(t))S(t)
−1
+/
x

n
¸
t=1
«(t) P(0. t) −r(0)

.
3.2 Valuation by diversification revisited 55
Here, the first term is zero if
«(t) =
J
x+t−1
/
x
=
(t−1)1
q
x
=
t−1
¡
x 1
q
x+t−1
.
which is exactly equal to the probability that a person aged x at time 0 dies
in the interval (t −1. t], and the second term is zero if
r(0) =
n
¸
t=1
(t−1)1
q
x
P(0. t). (3.19)
This fair price differs from the classical formulas in that the usual discount
factors have been replaced by zero coupon bond prices. Finally, we mention
that if the sum insured is payable immediately upon death and not at the
end of each year as suggested by Equation (3.19), we need a continuous
version of (3.19). For example, this can be obtained by considering small
time intervals and noting that, for small h,
th
q
x

t
¡
x
µ(x+t)h.
Thus, Equation (3.19) becomes
r(0) =

n
0
t
¡
x
µ(x+t)P(0. t) dt. (3.20)
In this situation, it is no longer possible to ensure that
˜
L =0, no matter how
many different bonds one buys. The reason for this phenomenon is that, in
principle, the payments can occur at any time, i.e. there are, in principle,
infinitely many possible payment times. However, we can move
˜
L arbitrarily
close to zero by choosing sufficiently many zero coupon bonds. Another
aspect is, of course, that in practice one would work with finitely many
different payment times (for example once every day, week or month), which
would amount to choosing some discretization of Equation (3.20).
3.2.7 Market values and zero coupon bonds
Consider now the main example from Chapter 2 with an endowment insurance
with a continuously payable premium r. Upon survival to n, the policy holder
receives the guaranteed amount l
a
(0), whereas l
ad
is payable immediately
upon a death before time n. We assume in addition that the contract starts at
time 0, where the policy holder is x years old. As in the previous chapter, we
assume that bonuses are used to increase the amount payable upon survival
only. Accordingly, we denote by l
a
(t) the amount guaranteed at time t, so
that l
a
(t) ≥l
a
(0).
56 Interest rate theory in insurance
The first order basis
Using the notation of Chapter 2, we can determine the premium r calculated
under the deterministic first order basis (r

. µ

) via
ro

xn
=l
a
(0)
n
E

x
+l
ad
A
1∗
xn
.
where
n−t
E

x+t
=exp

n
t
r

(t)dt

n−t
¡

x+t
.
o

x+tn−t
=

n
t
exp

s
t
r

(t)dt

s−t
¡

x+t
ds.
A
1∗
x+tn−t
=

n
t
exp

s
t
r

(t)dt

s−t
¡

x+t
µ

(x+s)ds.
and where the survival probability under the first order valuation principle is
given by
n−t
¡

x+t
=exp

n
t
µ

(x+t)dt

.
Similarly, we can give prospective expressions for the technical reserve under
the first order valuation principle. The technical reserve at time t for the
payments guaranteed at time t is given by
V

(t) =l
a
(t)
n−t
E

x+t
+l
ad
A
1∗
x+tn−t
−ro

x+tn−t
.
More generally, the technical reserve at time u for the payments guaranteed
at time t is given by
V

(t. u) =l
a
(t)
n−u
E

x+u
+l
ad
A
1∗
x+un−u
−ro

x+un−u
.
For u ≥t, we have the following differential equation with side conditions:
o
ou
V

(t. u) =r

(u) V

(t. u) +r−µ

(x+u)R

(t. u). (3.21)
V

(t. t) =V

(t).
V

(t. n) =l
a
(t).
where R

(t. u) is the sum at risk:
R

(t. u) =l
ad
−V

(t. u).
3.2 Valuation by diversification revisited 57
Market value of guaranteed payments
In Chapter 2 it was assumed that the mortality intensities µ and µ

and the
interest rates r and r

were deterministic functions. In fact, all formulas there
rely on this assumption. However, by repeating the combined diversification
and hedging argument of Section 3.2.6, it follows that the market value (the
fair price) at time t of the payments upon survival to n, guaranteed at time 0,
is given by
l
a
(0) P(t. n)
n−t
¡
x+t
.
where
n−t
¡
x+t
is the true survival probability. Similarly, the market value at
time t for the part of the contract which pays l
ad
immediately upon a death
before time n is given by
l
ad

n
t
P(t. s)
s−t
¡
x+t
µ(x+s) ds.
where P(t. s) is the price at time t of a zero coupon bond with expiration
time s; the market value at time t of the future premiums is given by
r

n
t
P(t. s)
s−t
¡
x+t
ds.
Note that the market value at time t of the future premiums is calculated in
the same way as the market value of the benefits. They both involve the zero
coupon bond prices at time t instead of the usual discount factors. From a
theoretical point of view, the only difference between the benefits and the
premiums is the sign. However, the situation becomes much more complicated
if one includes the policy holder’s possibilities for surrender and for changing
the contract into a free policy as indicated in Chapter 2. Combining the three
expressions above, we get the following expression for the market value at
time t for payments guaranteed at time 0:
V
g
(0. t) = l
a
(0) P(t. n)
n−t
¡
x+t
+

n
t
P(t. s)
s−t
¡
x+t
(µ(x+s)l
ad
−r) ds. (3.22)
We denote by V
g
(t) the market value at time t for the payments guaranteed
at time t. This quantity is obtained from Equation (3.22) by replacing l
a
(0)
by l
a
(t).
When the actual basis (r. µ) is deterministic, i.e. when r and µ are both
deterministic functions, Section 3.2.5 shows that the zero coupon bond prices
are given by
P(t. s) =exp

s
t
r(t) dt

. (3.23)
58 Interest rate theory in insurance
If we insert this expression into Equation (3.22), the market value at time u
for the payments guaranteed at time t can be written as follows:
V
g
(t. u) = l
a
(t) exp(−

n
u
r(t) dt)
n−u
¡
x+u
+

n
u
exp(−

s
u
r(t) dt)
s−u
¡
x+u
(µ(x+s)l
ad
−r) ds (3.24)
= l
a
(t)
n−u
E
x+u
+l
ad
A
1
x+un−u
−ro
x+un−u
. (3.25)
which corresponds to the formulas derived in Chapter 2.
What’s next?
The above calculations indicate that the market value of the guaranteed pay-
ments can be calculated by replacing the usual discount factors with zero
coupon bond prices. A natural question is therefore: does this solve the prob-
lem of determining market values of the guaranteed payments from a life
insurance contract completely? A quick answer is: yes! However, we insist
on continuing the discussion here for various reasons. One aspect is that zero
coupon bonds are not determined from the true market interest rate r via
Equation (3.23) if the market interest rate is stochastic. We can only use this
simple formula when the market interest rate is assumed to be deterministic or
even constant. In particular, this leads to the question of whether it is possible
to construct versions of Thiele’s differential equation for the market value
of the guaranteed payments. Here, the concept of forward rates proves to be
useful. Another aspect is that the majority of bonds on most bond markets are
more complicated than zero coupon bonds. This fact requires a more detailed
treatment of the relation between the prices of these more general bonds and
zero coupon bonds.
3.3 Zero coupon bonds and interest rate theory
In this section, we recall some fundamental concepts related to bond markets.
We address questions like: What is the term structure of interest rates? What
is a forward rate, and what is the difference between forward rates and the
market interest rate? What is credit risk? What is the role of these concepts
in the calculation of market values in life and pension insurance? We start
by analyzing zero coupon bonds, which can be viewed as the basic building
blocks of the bond market. In Section 3.5 below we discuss connections to
coupon bonds. In the following, T and T

are fixed (deterministic) finite
times.
3.3 Zero coupon bonds and interest rate theory 59
Definition 3.1 A zero coupon bond with maturity date T (also called a T-
bond) is a contract which pays one unit at time T. The price at time t ∈ |0. T]
is denoted by P(t. T ).
Throughout, we take P(t. t) =1.
Often the contracts defined above are called default free zero coupon
bonds. This serves to underline that the bonds cannot (or are very unlikely to)
default, i.e. that the issuers of the bonds are not likely to go bankrupt. This is
in contrast to so-called defaultable bonds, which are issued by companies (or
countries) which are less credit-worthy. As a consequence, bonds issued by
such parties typically give a higher return to the holders, since the bonds may
be worthless if the issuer goes bankrupt. This risk is also known as credit risk;
see Bielecki and Rutkowski (2001), Lando (2004) and Schönbucher (2003).
In the following, we focus on default free bonds.
3.3.1 Yield curves
Definition 3.2 The continuously compounded zero coupon yield (or the con-
tinuously compounded spot rate) R(t. T ) for the period |t. T ] is defined by
R(t. T) =−
1
T −t
logP(t. T).
It follows directly from this definition that the price of the zero coupon bond
can be expressed in terms of the continuously compounded yield as follows:
P(t. T ) =exp(−R(t. T )(T −t)) . (3.26)
so that Equation (3.26) can be interpreted as a discount factor obtained by
using the (constant) interest rate R(t. T ) during the interval |t. T ]. Note that
this (constant) interest rate R(t. T ) applies for any time u ∈ |t. T ], but that
the intensity depends on the price of the zero coupon bond at time t.
Definition 3.3 The term structure of interest rates at time t (or the zero
coupon yield curve) is given by the mapping
h →R(t. t +h).
3.3.2 Forward rates
Now consider times 0 ≤ t ≤ T

≤ T. We are interested in finding a deter-
ministic rate at time t (today) for a future investment made at time T

and
terminated at time T. This quantity is exactly the forward rate. We cannot
60 Interest rate theory in insurance
use the yield R(T

. T ) since this is defined in terms of P(T

. T ), which, in
general, is not known at time t. Instead, one introduces the concept of forward
rates.
Definition 3.4 The continuously compounded forward rate at time t for the
period |T

. T] is defined by
1(t. T

. T ) =−
logP(t. T ) −logP(t. T

)
T −T

. (3.27)
A better understanding of the concept of forward rates can be obtained
by noting that the ratio of the price at time t for a T-bond and a T

-bond is
given by
P(t. T )
P(t. T

)
=exp(−1(t. T

. T )(T −T

)). (3.28)
This can be given the following interpretation:
• at time t, sell one T

-bond and receive P(t. T

) (this corresponds to a loan);
• still at time t, use P(t. T

) to buy P(t. T

)¡P(t. T) units of T-bonds;
• at time T

pay one unit on the T

-bond (pay back the loan);
• at time T receive P(t. T

)¡P(t. T) from the T-bond.
The two transactions at time t are chosen such that the result at time t
is exactly 0, since the T-bonds are financed by selling a T

-bond. A result
of these transactions is that we have to pay (or invest) one unit at time T

.
At time T we receive the result of this investment, P(t. T

)¡P(t. T). Thus,
by trading zero coupon bonds at time t we are, in a sense, able to fix the
future return (or interest) for an amount to be invested at a future time T

.
Furthermore, from Equation (3.28) we obtain that
P(t. T

)
P(t. T)
=exp(1(t. T

. T)(T −T

)).
This shows that 1(t. T

. T) can be interpreted as the constant interest rate,
known at time t, which should be used for discounting future payments from
time T to time T

.
In our subsequent analysis in Section 3.3.5 of market values in life and
pension insurance, it is useful to work with the instantaneous forward rate
at time t for a given future time T, i.e. the limit of 1(t. T

. T) as T

T.
Equation (3.27) shows that this is simply the derivative of −logP(t. T) with
respect to T. So, provided that the mapping T → P(t. T) is continuously
differentiable with respect to the maturity date T, we can state the following
definition.
3.3 Zero coupon bonds and interest rate theory 61
Definition 3.5 The instantaneous forward rate at time t for time T is
defined by
1(t. T) =−
o
oT
logP(t. T). (3.29)
The instantaneous short rate at time t is r(t) =1(t. t).
We show in the following how forward rates appear in a natural way in
Thiele’s differential equation for the market value of the guaranteed payments
from a standard life insurance contract. It follows immediately from the
definition above that

T
t
1(t. t) dt =−logP(t. T) +logP(t. t).
Since P(t. t) =1, we see that the value at time t of the zero coupon bond is
given by
P(t. T) =exp

T
t
1(t. t) dt

. (3.30)
Note, however, that we cannot in general conclude that the zero coupon
bond prize is of the form P(t. T) = exp(−

T
t
r(u) du). To obtain this result,
additional assumptions are required. More precisely, it is necessary to assume
that future bond prices are known at time t, which basically means that the
instantaneous short rate is deterministic. This situation was considered in
Section 3.2.5.
Equation (3.30) shows that the zero coupon bond prices can in fact be
derived directly from the instantaneous forward rates. Thus, there are several
possibilities for specifying models for a bond market:
• specify all P(t. T) for 0 ≤ t ≤ T ≤ T

, where T

is some maximum time
point;
• specify all 1(t. T) for 0 ≤t ≤T ≤T

and derive P(t. T) via Equation (3.30)
and the instantaneous short rate r(t) via Definition 3.5;
• specify the development of r(t), for 0 ≤t ≤T ≤T

.
We will not go into a treatment of the mathematical aspects associated
with these different methods. However, we mention that one has to be careful
when using the first two approaches in order to avoid arbitrage possibilities.
The last approach also requires some additional work, since it is not clear
how r determines the zero coupon bond prices.
62 Interest rate theory in insurance
3.3.3 Relations between forward rates and spot rates
We have now introduced several quantities, such as zero coupon bond, yield,
spot rate and forward rate, and the reader may wonder if all this is really nec-
essary and how all these quantities are related. Some immediate consequences
of the above definitions are as follows.
(1) From Equation (3.28) with T

=t, it follows that
1(t. t. T) =R(t. T).
i.e. the forward rate at time t for |t. T] coincides with the spot rate for
|t. T].
(2) Combining Equations (3.26) and (3.30), we see that
1
T −t

T
t
1(t. t) dt =R(t. T). (3.31)
which shows that the zero coupon yield for |t. T] can be interpreted as
the average of the instantaneous forward rates.
3.3.4 Simple rates
As an alternative to the continuously compounded rates introduced above, one
can also introduce simple yields (or simple spot rates) and simple forward
rates. For completeness, we give a short discussion of this concept, which
does not play a further role, however, in our treatment of market values in life
and pension insurance in the present chapter. They reappear in our discussion
of swap rates in Chapter 7.
Simple rates differ from the usual principle of compounding rates in the
sense that past interest is not included in the calculation of the interest for a
given period. If we deposit an amount P at time 0 in an account with simple
rate L (per time unit), this leads to the interest PL(t −s) for the time period
|s. t]. With continuously compounding interest under constant interest rate r,
the deposit on the savings account would increase from Pe
rs
at time s to Pe
rt
at time t. Thus, the interest credited during the interval |s. t] is given by
P(e
rt
−e
rs
) =Pe
rs
(e
r(t−s)
−1).
This can now be formalized via the following definition.
3.3 Zero coupon bonds and interest rate theory 63
Definition 3.6 The simple yield (or the simple spot rate or even the LIBOR
spot rate) for |t. T] is given by
L(t. T) =−
P(t. T) −1
(T −t)P(t. T)
.
and the simple forward rate (or the LIBOR forward rate) for |T

. T] at time
t is given by
L(t. T

. T) =−
P(t. T) −P(t. T

)
(T −T

)P(t. T)
.
It follows from the definition of the simple spot rate that
L(t. T)(T −t)P(t. T) =1−P(t. T).
which has the following interpretation: 1−P(t. T) is the return or gain during
the interval |t. T] from buying the T-bond at time t at the price P(t. T) and
cashing the amount 1 at time T. Under the principle of simple interest, the
amount
P(t. T)L(t. T)(T −t)
is exactly the interest which accrues in the interval |t. T] in connection with
the investment of P(t. T) under a constant simple interest L(t. T).
3.3.5 Market values and forward rates
In this section we derive a version of Thiele’s differential equation for the
market values of the guaranteed payments that involves forward rates. In
Section 3.3.2 we showed how the instantaneous forward rates 1(t. t) at time
t are related to the price of a zero coupon bond at time t; see Equation (3.30).
If we insert this expression into Equation (3.25) for the market value at
time u ≥t for the payments guaranteed at time t, we immediately obtain the
following expression:
V
g
(t. u) =l
a
(t) exp

n
u
1(u. t) dt

n−u
¡
x+u
+

n
u
exp

s
u
1(u. t) dt

s−u
¡
x+u
(µ(x+s)l
ad
−r) ds. (3.32)
The situation t = u gives the market value V
g
(t) at time t of the payments
guaranteed at time t. This auxiliary quantity is of importance in situations
with stochastic interest, where Equation (3.25) cannot be applied directly.
In this situation, we can derive an alternative expression for the market value
64 Interest rate theory in insurance
which is similar to the classical formulas, but where the stochastic interest
rate is replaced by the instantaneous forward rates at time u, which are known
at time u.
We give here a version of Thiele’s differential equation, which also opens
up the possibility of deriving a differential equation for the individual bonus
potential in a model with stochastic interest rates. It is not possible to derive a
differential equation for V
g
(t. u) by applying Equation (3.32), since V
g
(t. u)
is a rather complicated function of u which moreover depends on the interest
rate at time u. Instead, we introduce the additional auxiliary function
V
g.
(t. u) =l
a
(t) exp

n
u
1(t. t) dt

n−u
¡
x+u
+

n
u
exp

s
u
1(t. t) dt

s−u
¡
x+u
(µ(x+s)l
ad
−r) ds.
(3.33)
which may be interpreted as a value at time u for the payments guaranteed at
time t, calculated by applying the instantaneous forward rates at time t. Thus,
V
g.
(t. u) is not really a market value. However, the main advantage of this
quantity is that it is a very simple function of u, which can be differentiated
directly. Hence we obtain the following differential equation for u ≥ t with
side conditions:
o
ou
V
g.
(t. u) =1(t. u)V
g.
(t. u) +r−µ(x+u)R
g.
(t. u). (3.34)
V
g.
(t. t) =V
g
(t).
V
g.
(t. n) =l
a
(t).
where R
g.
is the sum at risk, given by
R
g.
(t. u) =l
ad
−V
g.
(t. u).
This differential equation provides an alternative method for calculating Equa-
tion (3.33) as follows. Given the sum insured l
a
(t) guaranteed at time t and
the forward rates 1(t. u) at time t, the market value V
g
(t) can be calculated
by solving Equation (3.34) on the interval |t. n] with the terminal condition
corresponding to the payment upon survival to n, guaranteed at time t. For
a description of the methods for solving differential equations numerically,
see, for example, Schwarz (1989).
The individual bonus potential
The forward rates can also be used for deriving an expression for the so-called
individual bonus potential V
ib
. The most simple situation is the case where
3.4 A numerical example 65
V(t) ≥ V

(t) ≥ V
g
(t), where V(t) is the total reserve for the contract; see
Chapter 2. In this case, the individual bonus potential is given by
V
ib
(t) =V

(t) −V
g
(t).
By applying the differential equations, Equations (3.21) and (3.34), together
with the terminal condition V

(t. n) −V
g.
(t. n) =l
a
(t) −l
a
(t) =0, we find
that
V
ib
(t) =

n
t
exp

s
t
(1(t. t) +µ(x+t)) dt

c(t. s) ds. (3.35)
where
c(t. s) =(1(t. s) −r

(s))V

(t. s) +(µ

(x+s) −µ(x+s))R

(t. s). (3.36)
Equations (3.35) and (3.36) differ from the corresponding expressions in
Chapter 2 (see Equations (2.28) and (2.29)), in that the market interest rate has
been replaced by the forward rates. This serves to underline the importance of
the forward rates in the situation where the interest rate is stochastic, since the
safety loadings c(t. s) for the payments guaranteed at time t now involve the
forward rates at time t. The expression in Equation (3.35) has the same
interpretation as Equation (2.28) in Chapter 2: the individual bonus potential
is the market value of the safety loadings on the guaranteed payments.
3.4 A numerical example
In this section, we illustrate the role of the bonus potentials via a numerical
example. We consider an insurance contract, where the benefits have been
calculated using a first order mortality intensity of Gompertz–Makeham form
as follows:
µ(x+t) =o+8c
x+t
. (3.37)
The actual parameters used are listed in Table 3.1. This table also contains
Gompertz–Makeham estimates for the Danish population taken from Dahl
and Møller (2006) for males and females for 1980 and 2003. The estimated
mortality intensities can be found in Figures 3.1 and 3.2, together with the
estimates for 1970 and 1990. The estimated mortality intensities are clearly
ordered and reveal a decreasing trend from 1970 to 2003. This decrease in the
mortality intensity implies that the expected lifetimes have increased during
the same period. Figures 3.3 and 3.4 show the expected lifetimes for age 30
and age 65; these figures are also taken from Dahl and Møller (2006).
66 Interest rate theory in insurance
Table 3.1. The first order mortality and estimated Gompertz–Makeham
parameters for 1980 and 2003.
Males Females
o 8 c o 8 c
First order 0.0005 0.00007586 1.09144 0.0005 0.000053456 1.09144
1980 0.000233 0.0000658 1.0959 0.000220 0.0000197 1.1063
2003 0.000134 0.0000353 1.1020 0.000080 0.0000163 1.1074
30 40 50 60 70 80
0.00
0.10
0.20
Figure 3.1. Estimated mortality intensities for males. Solid lines are 1970
estimates; dashed lines correspond to 1980; dotted lines to 1990; and dot-dashed
lines to 2003.
30 40 50 60 70 80
0.00
0.10
Figure 3.2. Estimated mortality intensities for females. Solid lines are 1970
estimates; dashed lines correspond to 1980; dotted lines to 1990; and dot-dashed
lines to 2003.
Figure 3.3 shows that the remaining lifetimes have increased by approxi-
mately 2.5 years for males and 1.5 years for females aged 30 from 1980 to
2003. Using this method, the expected lifetime in 2003 is about 75.3 years
for 30-year-old males and 79.5 for 30-year-old females. A closer study of
the underlying Gompertz–Makeham parameters (o. 8. c) indicates that o has
decreased during the period from 1960 to 2003 for both males and females.
The estimates for 8 increase from 1960 to 1990, whereas the estimates for
c decrease. In contrast, 8 decreases and c increases in the last part of the
period considered (from 1990 to 2003). Table 3.2 shows the expected lifetimes
3.4 A numerical example 67
Table 3.2. Expected lifetimes with the first order basis and the market-value
bases for 1980 and 2003.
Age First order Est. 1980 Est. 2003 Est. 2003 Est. 2003
(longevity 0.5%) (longevity 1%)
Males 30 74.1 73.1 75.8 77.8 79.9
65 78.5 78.9 80.3 80.8 81.2
90 93.7 93.1 93.4 93.4 93.4
Females 30 77.9 78.8 80.1 82.2 84.5
65 81.5 82.4 83.3 83.8 84.4
90 94.9 94.0 94.3 94.3 94.4
1960 1970 1980 1990 2000
70
74
78
Figure 3.3. Development of the expected lifetime of 30-year-old females (dotted
line) and 30-year-old males (solid line) from 1960 to 2003. Estimates are based
on the last five years of data available at the given years.
1960 1970 1980 1990 2000
76
80
84
Figure 3.4. Development of the expected lifetime of 65-year-old females (dotted
line) and 65-year-old males (solid line) from 1960 to 2003.
calculated for the first order basis and the estimated mortalities for age 30, 65
and 90. With the 2003 estimate, moreover, we have included simple correc-
tions for longevity, i.e. for future reductions in the mortality, by assuming that
the mortality in all ages decreases each year by 0.5% and 1%, respectively.
In this example, we see that the expected lifetimes calculated under the first
order basis are smaller than the expected lifetimes calculated under the 2003
estimate. For males and females aged 30 or 65, the difference is between
1.7 and 2.2 years. If we include a correction for future improvements in the
68 Interest rate theory in insurance
1960 1970 1980 1990 2000
4 × 10
–4
1 × 10
–4
(a)
6 × 10
–5
2 × 10
–5
1960 1970 1980 1990 2000
(b)
1960 1970 1980 1990 2000
(c)
1.095
1.105
1.115
Figure 3.5. Development of the Gompertz–Makeham parameters: (a) o; (b)
8; (c) c for females (dotted lines) and males (solid lines) from 1960 to 2003.
Estimates are based on the last five years of data available at the given years.
mortality, we see that this difference increases considerably, in particular at
low ages. Using a yearly decline of 1%, we see that this difference is now
between 2.7 and 6.6 years.
In the example, we have determined market values by using the above
2003 mortality intensity with a yearly 1% correction for future mortality
improvements. In addition, we have applied a zero coupon yield curve from
31 December 2005 for discounting the payments. These numbers, which have
essentially been adjusted for tax by multiplying the observed yield curve by
0.85, can be found in Table 3.3 for maturity t =1. 2. . . . . 30; for t ≥30, we
use the value for maturity 30. We have used this yield curve for the calculation
of all market values below. The expected present values used in the example
can be found in Table 3.4 for first order bases with (approximate) interest
3.4 A numerical example 69
Table 3.3. Zero coupon yields from 31 December 2005 before and after tax.
Maturity Yield Yield after tax Maturity Yield Yield after tax
1 0.02843 0.02422 16 0.03638 0.03100
2 0.02991 0.02548 17 0.03660 0.03120
3 0.03084 0.02627 18 0.03683 0.03139
4 0.03090 0.02632 19 0.03706 0.03159
5 0.03114 0.02653 20 0.03729 0.03178
6 0.03162 0.02694 21 0.03733 0.03182
7 0.03203 0.02729 22 0.03737 0.03185
8 0.03277 0.02792 23 0.03741 0.03189
9 0.03365 0.02867 24 0.03746 0.03193
10 0.03404 0.02901 25 0.03750 0.03196
11 0.03447 0.02937 26 0.03754 0.03200
12 0.03489 0.02973 27 0.03758 0.03203
13 0.03531 0.03009 28 0.03763 0.03207
14 0.03573 0.03045 29 0.03767 0.03211
15 0.03615 0.03081 30 0.03771 0.03214
rates 4.5%, 2.5% and 1.5%, respectively. This table also contains the expected
present value calculated by using the zero coupon yield curve of Table 3.3
and the estimated mortality from 2003 with a yearly longevity correction of
1%. We consider now three contracts signed at age 30 with first order interest
rate 4.5%, 2.5% and 1.5%, respectively. We set the life annuity payment rate
to unity and sum paid at death to five. Premiums are paid continuously at a
fixed rate as long as the policy holders are alive and stop at the retirement age
of 65. For each of these three contracts, we have determined the equivalence
premium at age 30, such that the technical reserves start at zero at time 0.
We have then calculated the technical reserve at various ages by rolling
forward the policy holders’ accounts with the first order assumptions. The
fair premiums can be found in Table 3.5 for the three contracts with different
technical interest rates. This table shows (not surprisingly) that the premiums
and technical reserves increase if the technical interest decreases. (For a
study of the dependence of the technical reserve on the interest rate and other
parameters, see Kalashnikov and Norberg (2003).) Table 3.5 also includes
the so-called free policy factors introduced in Section 2.7.1. These factors are
used for calculating the market value V
f
of the guaranteed free policy benefits.
We see that the factors decrease when the technical interest is decreased.
The relevant market values can be found in the lower part of Table 3.5
for the three contracts. At the time of signing these contracts at age 30, we
see that the market values of the guaranteed payments are negative for the
two contracts with low technical interest rates (2.5% and 1.5%); this leads to
70 Interest rate theory in insurance
Table 3.4. Expected present values (for males) under the first order
valuation basis and the market-value basis for the premiums, term
insurance, pure endowment and the life annuity.
Premiums are paid continuously and are assumed to stop at age 65, where the
term insurance and pure endowment expire. The life annuity is assumed to start
at age 65.
Age Premiums Term insurance Pure endowment Life annuity
First order
interest 4.5%
30 17.011 0.086 0.165 1.689
40 14.273 0.110 0.262 2.681
50 10.281 0.122 0.425 4.356
65 0.000 0.000 1.000 10.239
90 0.000 0.000 0.000 3.277
First order
interest 2.5%
30 22.069 0.130 0.323 3.880
40 17.406 0.145 0.423 5.080
50 11.679 0.143 0.568 6.810
65 0.000 0.000 1.000 11.997
90 0.000 0.000 0.000 3.460
First order
interest 1.5%
30 25.519 0.161 0.455 5.952
40 19.384 0.168 0.540 7.069
50 12.492 0.155 0.657 8.595
65 0.000 0.000 1.000 13.080
90 0.000 0.000 0.000 3.560
Market values,
1% longevity
30 20.709 0.068 0.278 3.753
40 16.804 0.090 0.384 4.994
50 11.595 0.103 0.546 6.798
65 0.000 0.000 1.000 12.188
90 0.000 0.000 0.000 3.200
bonus potentials at the time of signing the contract. In contrast, the market
value is greater than zero at time 0 for the contract with technical interest
4.5%, which implies that the company needs to set aside additional capital
for this contract. For the contracts with technical interest 2.5% and 1.5%, we
see that the market values of the guaranteed free policy benefits are typically
higher than the market values for the guaranteed payments. For the contract
with interest rate 4.5%, this is not the case. The individual bonus potentials
for this example are listed finally in Table 3.6. We see that the contract with
the high technical interest does not lead to any bonus potentials. (At very high
ages, however, small bonus potentials arise because the first order mortality
is lower than the estimated mortality at high ages.) For the two contracts with
technical interest 1.5% and 2.5%, the bonus potentials on the future premiums
3.5 Bonds, interest and duration 71
Table 3.5. Fair premiums (equivalence premiums fixed at age 30), technical
reserves and free policy factors for the three policies with first order
interest 4.5%, 2.5% and 1.5%, respectively (upper part). The market values
of guaranteed payments (GP) and market values of guaranteed free policy
payments are given in the lower part.
Contract: term insurance of five units and life annuity of one unit, with continuously
paid premiums.
Premiums 0.125 0.205 0.265
Interest 4.50% 2.50% 1.50% 4.50% 2.50% 1.50%
Age Technical reserves Free policy factors
30 0.000 0.000 0.000 0.000 0.000 0.000
40 1.451 2.235 2.776 0.449 0.385 0.351
50 3.685 5.128 6.062 0.742 0.682 0.647
65 10.239 11.997 13.080 1.000 1.000 1.000
90 3.277 3.460 3.560 1.000 1.000 1.000
Age Market values, GP Market values, free policy
30 1.511 −0.157 −1.393 0.000 0.000 0.000
40 3.347 1.994 0.991 2.445 2.095 1.910
50 5.868 4.934 4.242 5.426 4.984 4.731
65 12.188 12.188 12.188 12.188 12.188 12.188
90 3.200 3.200 3.200 3.200 3.200 3.200
decrease as the age approaches the age of retirement. In contrast, the bonus
potentials on the free policy start at zero and increase with the premiums paid.
However, at some point, they start to decrease again and eventually vanish
as the age increases.
3.5 Bonds, interest and duration
The majority of bonds traded in most bond markets are more general than
zero coupon bonds. Typical examples are annuity bonds and bullet bonds.
Here, we give a framework that can be used for basically any bond with
predetermined payments, which shows how the prices of these more general
bonds are related to the prices of zero coupon bonds.
3.5.1 A general bond
Consider a bond issued at time t
0
, say, which specifies payments c
1
. . . . . c
n
at given times t
1
- · · · - t
n
. The value of such a bond at time t ≥ t
0
(after
72 Interest rate theory in insurance
Table 3.6. The total liability, the additional reserve needed and the two
individual bonus potentials (BP): on premiums and on the free policy.
Interest 4.50% 2.50% 1.50% 4.50% 2.50% 1.50%
Age Total liability Additional reserve included
30 1.511 0.000 0.000 1.511 0.000 0.000
40 3.347 2.235 2.776 1.896 0.000 0.000
50 5.868 5.128 6.062 2.183 0.000 0.000
65 12.188 12.188 13.080 1.949 0.191 0.000
90 3.277 3.460 3.560 0.000 0.000 0.000
Age Bonus potential, free policy Bonus potential, premiums
30 0.000 0.000 0.000 0.000 0.157 1.393
40 0.000 0.140 0.866 0.000 0.100 0.919
50 0.000 0.144 1.331 0.000 0.050 0.489
65 0.000 0.000 0.892 0.000 0.000 0.000
90 0.077 0.260 0.360 0.000 0.000 0.000
possible payments at t) is given by
P(t) =
¸
i:t
i
>t
P(t. t
i
)c
i
. (3.38)
where P(t. t
i
) is the price at time t of a zero coupon bond with maturity t
i
.
Again, a simple arbitrage argument shows that this is the only price that does
not lead to the possibility of generating risk-free gains in a market where it is
possible to buy and sell the bond in Equation (3.38) as well as zero coupon
bonds with maturities t
1
. . . . . t
n
. This can be seen by noting that payment
streams of the form c
k
. . . . . c
n
at times t
k
- · · · - t
n
can be generated by
buying exactly c
k
t
k
-bonds, c
k+1
t
k+1
-bonds, etc.
Example 3.7 (Annuity bond) Taking c
k
=c for k =1. . . . . n gives an annuity
bond.
Example 3.8 (Bullet bond) Fix some simple rate L and some amount K (the
principal). Consider the situation where
c
k
=L(t
k
−t
k−1
)K.
for k =1. . . . . n−1 and
c
n
=L(t
n
−t
n−1
)K+K.
3.5 Bonds, interest and duration 73
With this bond, the holder receives at time t
k
the simple interest L(t
k
−t
k−1
)
on the principal K for the interval |t
k−1
. t
k
]. At time t
n
, the principal K is
paid back together with interest for the interval |t
n−1
. t
n
].
3.5.2 Yield and duration
In Section 3.3.1, we introduced the zero coupon yield R(t. T), which was
defined as the constant intensity for |t. T] which allows the zero coupon bond
with maturity T to be expressed as a traditional discounting factor of the
following form:
P(t. T ) =exp(−R(t. T)(T −t)).
A natural generalization of this concept can be given for more general bonds.
Here, the yield to maturity x(t) is the constant rate which ensures that the
discounted value of all future payments corresponds to the value at time t of
the bond. This can be made more precise via the following definition.
Definition 3.9 The yield to maturity at time t of the bond in Equation (3.38)
is the solution x(t) to the following equation:
P(t ) =
¸
i:t
i
>t
e
−x(t)(t
i
−t)
c
i
=: E(t. x(t)). (3.39)
Note that if all c
i
are non-negative and c
i
>0 for some i with t
i
>t, then
the mapping x(t) →E(t. x(t)) is strictly decreasing, and hence there exists
a unique non-negative solution to Equation (3.39). Thus, x(t) is the constant
rate of return that the holder receives for the payments during (t. t
n
].
A natural question is now the following: how sensitive is the price P(t) of
the bond to changes in the yield to maturity? Or, more precisely, how will
a change in the yield to maturity affect the bond price? It can be relevant
to investigate, for example, how sensitive the value of the company’s total
portfolio of bonds is to changes in the term structure. Similarly, one might be
interested in a comparison of the sensitivity of different bonds. The answers
to these questions are closely related to the so-called duration of the bond,
which can be interpreted as the weighted average of the time to the payments
c
i
≥0 weighted by the factors c
i
e
−x(t)(t
i
−t)
. These factors represent the values
at time t of the payments c
i
discounted by using the yield x(t). The precise
definition of the concept of duration is given here.
74 Interest rate theory in insurance
Definition 3.10 The duration D(t) at time t of the bond in Equation (3.38)
with yield to maturity x(t) is defined by
D(t) =
¸
i:t
i
>t
(t
i
−t) e
−x(t)(t
i
−t)
c
i
P(t)
. (3.40)
We see from this definition that the duration at time t for a zero coupon
bond with maturity t is exactly equal to the time to maturity t −t. If we
think of P(t) in Equation (3.39) as a function of the yield to maturity x(t), we
obtain, by differentiating Equation (3.39) with respect to x(t), the following:
d
dx(t)
P(t) =−
¸
i:t
i
>t
(t
i
−t) e
−x(t)(t
i
−t)
c
i
=−P(t) D(t). (3.41)
For more details on duration, see, for example, Hull (2005) and references
therein.
3.6 On the estimation of forward rates
In practical situations, one is often confronted with the following problem.
Given that we have observed prices of a number of bonds, what can be said
about the zero coupon yield curve and the forward rates? It is important to
note that so far we have not introduced any model to describe how the price
P(t. T) of a given zero coupon bond evolves over time. This will be done
later. At this point, we simply think of the prices as given and try to extract
as much information from the observations as possible without introducing
complicated models.
We consider the most simple situation where we observe the prices at time
t of some zero coupon bonds and address the problem of deriving the forward
rate curve.
3.6.1 Observing prices of zero coupon bonds
We fix times t
1
- · · · - t
n
and assume that we have observed at time t the
prices P(t. t
i
) of zero coupon bonds with maturities t
i
, for i ∈ 1(t), where
1(t) =]; ∈ ]1. . . . . n]t
;
>t].
i.e. of bonds that have not expired at time t. (We use this notation in order
to emphasize the dynamic nature of this problem. Alternatively, we could
simply consider one fixed time, t = 0, say, and work with a fixed set I of
observed bond prices with maturities t
1
. . . . . t
n
.) We can now derive the zero
3.6 On the estimation of forward rates 75
coupon yield for the periods |t. t
i
], i ∈ 1(t), by use of Definition 3.2, which
shows that
R(t. t
i
) =−
1
t
i
−t
logP(t. t
i
). for i ∈ 1(t).
One can visualize this by plotting the points (t
i
−t. R(t. t
i
)), i ∈ 1(t), which
are the pairs of time to maturity and zero coupon yield for the corresponding
period.
Definition 3.4 determines the forward rates for the periods |t
i−1
. t
i
],
given by
1(t. t
i−1
. t
i
) =−
logP(t. t
i
) −logP(t. t
i−1
)
t
i
−t
i−1
.
However, we are also interested in (an estimate for) the entire zero coupon
yield curve h →R(t. t +h) at time t or the curve for the instantaneous forward
rates h →1(t. t +h).
One typical approach to the first problem is to apply some numerical
technique to fit the best smooth curve to the observed points. Alternatively,
one can fix some parameterized family of possible instantaneous forward rate
curves,
]h →1(t. t +h; 0)0 ∈ O].
and fit this curve to the estimated zero coupon yields R(t. t
i
), i ∈ 1(t), by
using Equation (3.31), which shows that
R(t. t
i
; 0) =
1
t
i
−t

t
i
t
1(t. u; 0) du.
Thus, the problem consists in finding the curve R(t. t +h; 0) which pro-
vides the best fit to the observed quantities R(t. t
i
), i ∈ 1(t), by using some
subjective criterion. This is discussed in more detail below.
3.6.2 The Nelson–Siegel parameterization
In this section, we review the so-called Nelson–Siegel parameterization for
the forward rates, suggested by Nelson and Siegel (1987); for an extension and
an application to the Swedish Treasury bills and coupon bonds, see Svensson
(1995). For simplicity, we fix t =0. Nelson and Siegel (1987) advocated the
following parametric family for the forward rates:
1(0. t) =8
0
+8
1
e
−t¡o
+8
2
t
o
e
−t¡o
. (3.42)
76 Interest rate theory in insurance
which is parameterized by 8 =(8
0
. 8
1
. 8
2
) and o. The first part, 8
0
, can be
interpreted as a long term component, the part with 8
1
is determining the short
term forward rate, and the 8
2
-component is affecting the medium term forward
rate. Finally, the parameter o is determining how “short term,” “medium term”
and “long term” should be defined. Integrating Equation (3.42) with respect
to the maturity t, we find that the zero coupon yield is given by
R(0. t) =
1
t

t
0
1(0. t) dt =8
0
+(8
1
+8
2
)(1−e
−t¡o
)
1
t¡o
−8
2
e
−t¡o
. (3.43)
The different components determining the forward rates are depicted in
Figure 3.6.
We now discuss a simple way of fitting Equation (3.43) to some observed
zero coupon yields for the periods |0. t
1
]. . . . . |0. t
n
]. In order to simplify
notation, we first change the parameterization slightly and introduce the
parameter 6 =(6
1
. 6
2
. 6
3
)
tr
, where tr means “transposed,” and
R(0. t
i
; 6. o) =6
1
+6
2
(1−e
−t
i
¡o
)
1
t
i
¡o
+6
3
e
−t
i
¡o
=: Y
i
(o)
tr
6. (3.44)
where
Y
i
(o) =

1. (1−e
−t
i
¡o
)
1
t
i
¡o
. e
−t
i
¡o

tr
.
Thus, for given o, R(0. t
i
) is linear in 6. This special property can be exploited
to yield a very simple approach to fitting the model to observed prices using
10 8 6 4 2 0
0.0
0.2
0.4
0.6
0.8
1.0
maturity
f
o
r
w
a
r
d

r
a
t
e
s
Figure 3.6. Nelson–Siegel parameterization: components determining the for-
ward rates. The horizontal line gives the level for “long term” forward rates; the
dashed line affects “medium term” rates, and the exponentially decreasing line
affects “short term” forward rates.
3.6 On the estimation of forward rates 77
the generalized least squares method. This approach can, for example, also be
used to fit a Gompertz–Makeham mortality intensity to occurrence-exposure
rates; see, for example, Norberg (2000).
Assume that we have observed zero coupon yields
ˆ
R(0. t
i
), for i =1. . . . . n.
(For example, these could be derived from zero coupon bond prices or from
some more general bonds as described above.) We then introduce the vector
of observed yields
ˆ
R =(
ˆ
R(0. t
1
). . . . .
ˆ
R(0. t
n
))
tr
and the (n×3)-matrix Y(o)
given by
Y(o) =



Y
1
(o)
tr
.
.
.
Y
n
(o)
tr



=




1 (1−e
−t
1
¡o
)
1
t
1
¡o
e
−t
1
¡o
.
.
.
.
.
.
.
.
.
1 (1−e
−t
n
¡o
)
1
t
n
¡o
e
−t
n
¡o




.
One possible way of fitting Equation (3.44) to the observed prices is now to
minimize over 6, for fixed o, the quadratic function
h(6) =
1
2
(Y(o)6−
ˆ
R)
tr
A(Y(o)6−
ˆ
R). (3.45)
where A is some symmetric, positive definite weight matrix. For exam-
ple, one could choose A equal to the (n×n)-identity matrix. Alternatively,
one could specify A in such a way that the estimated curve is forced to
be close to some specific yields; for example, one could use trading vol-
ume as a measure or choose to give more weight to yields from specific
maturities.
Equation (3.45) is minimized for 6 solving (o¡o6)h(6) =0, i.e.
Y(o)
tr
A(Y(o)6−
ˆ
R) =0. (3.46)
which has the following solution:
6(o) =(Y(o)
tr
AY(o))
−1
Y(o)
tr
A
ˆ
R. (3.47)
The optimal choice of o can then be found by inserting Equation (3.47) into
Equation (3.45) and minimizing h(6(o)) over o, i.e.
min
o
h(6(o)).
For more details on this approach applied for the estimation of mortality
intensities, see Norberg (2000); this method is also applied by Cairns (2004)
for the estimation of the yield curve.
78 Interest rate theory in insurance
3.7 Arbitrage-free pricing in discrete time
What is arbitrage-free pricing? What is hedging? What is a martingale mea-
sure? How can these concepts be used for the calculation of market values in
life and pension insurance?
This section provides an introduction to the principle of no arbitrage. The
main idea is that prices of traded assets should be determined in such a way
that no risk-less gains arise from trading these assets. The term asset is used
for basically any traded securities, such as bonds, stocks and even savings
accounts.
The theory presented in this section is particularly useful for explaining
how so-called derivatives can be priced. Important examples for life insur-
ance companies are options, swaptions and interest guarantee options. More
precisely, it can be shown how these instruments should be priced (within a
given model) in order to avoid the possibility of generating risk-free gains.
In addition, we see in Section 3.9 how these results can be applied when
calculating market values in life and pension insurance.
This section is organized as follows. Section 3.7.1 describes the market of
traded assets. In Section 3.7.2 we consider a simple two-period example with
two assets, where the interest rate for the last period can attain two different
values. Some fundamental results from probability theory that are crucial
for a more systematic introduction to arbitrage-free pricing are reviewed in
Section A.1 of the Appendix. Readers not familiar with concepts such as
filtrations, stochastic processes and martingales may find it helpful to consult
this appendix for more details.
Example 3.11 (An arbitrage) As an example of a risk-less gain or an arbi-
trage, consider the situation where there exist two traded assets S
0
and S
1
whose prices at time 0 coincide, that is S
0
(0) = S
1
(0). Assume, furthermore,
that we know that with probability 1 (almost surely; see the Appendix, Sec-
tion A.1) S
0
(T) ≥ S
1
(T) and that P(S
0
(T) > S
1
(T)) > 0. How can this be
exploited to make a risk-free gain? Well, it seems intuitively reasonable that
there is something wrong with this model, since asset number 0 seems to be to
cheap: it has the same price as asset number 1 at time 0, but it will have at
least the same value at time T. This observation can be exploited in the follow-
ing way. At time 0 buy one unit of asset number 0 and sell one unit of asset
number 1. (Thus, we assume that we can buy and sell the assets at the given
prices, and hence we are not dealing with transaction costs.) At time T this
leads to the gain V = S
0
(T) −S
1
(T) ≥ 0. By the above assumptions, we have
P(V ≥ 0) = 1 and P(V > 0) > 0, which has the following consequences. At
3.7 Arbitrage-free pricing in discrete time 79
time T we receive V ≥0 without having paid anything at time 0. Basically, this
means that we have received a lottery ticket for free.
The main assertion in arbitrage-free pricing is that no arbitrage possibilities
(like the one in Example 3.11) should exist. If they existed, all investors would
be interested in following these strategies, and this would imply that there
would only be buyers for the cheap asset (asset 0 in the example) and sellers
for the expensive asset (asset 1 in the example).
3.7.1 Traded assets and information
We consider a financial market consisting of two traded assets, whose prices
at time t are given by S
0
(t) and S
1
(t), respectively. (It is not difficult to
generalize the model to the case of J traded assets; however, this would only
serve to complicate notation and the presentation in general.) Here we depart
from the previous assumption of deterministic prices as in Section 3.2.5,
and allow for randomness so that future prices are not known in general.
More precisely, the prices at time u are described by the random variables
(S
0
(u). S
1
(u)), which are not observed before time u.
The family of prices for asset i in discrete time, S
i
= (S
i
(t))
t∈]0.1.... .T]
, is
an example of a stochastic process. To keep track of what is known at time t,
we introduce a so-called filtration F = ((t))
t∈]0.1.... .T]
. Here, (t) is the
information available at time t. We assume that the amount of information
is non-decreasing over time (we do not forget), so that (t) ⊆ (u) for
any t - u ≤ T. (Mathematically, a filtration is an increasing sequence of
u-algebras.) The price process S
i
is said to be adapted to the filtration F
if the price of asset i at time t is part of the information (t) available at
time t. (Mathematically this means that S
i
(t) is (t)-measurable.) The model
is illustrated in Figure 3.7.
A stochastic process is called predictable if its value at time t is already
known at time t −1. Thus, any predictable process is adapted.
prices: S
i
(0) S
i
(1) S
i
(t − 1) S
i
(t) S
i
(T )
information:
(0) (1) (t − 1) (t) (T )
time: 0 1 t − 1 t T
Figure 3.7. The price processes and the amount of information available.
80 Interest rate theory in insurance
We assume that S
0
describes the development of a savings account with
periodic interest i(t) for the period (t −1. t], so that
S
0
(t) =(1+i(1)) · · · (1+i(t)).
Typically, (i(t))
t∈]1.... .T]
is also a stochastic process.
For example, one can think of S
1
as the value of a zero coupon bond. As
in the previous sections, we apply the process S
0
as a discounting factor (also
called a numeraire) and we also introduce the discounted price processes X
and X
0
, defined by
X(t) =S
1
(t)¡S
0
(t) and X
0
(t) =S
0
(t)¡S
0
(t) =1.
3.7.2 A two-period example
Before we continue our discussion of arbitrage-free pricing, we consider a
simple example with two periods. We assume that trading takes place at
times t = 0. 1. 2. Assume that the interest i (1) for the savings account for
the first period is known at time 0, whereas the interest i(2) for the second
period is unknown at time 0 and is only revealed at time 1. The value of
the savings account at time 0 is given by unity, i.e. S
0
(0) = 1. Similarly,
S
0
(1) = (1 +i (1)) = e
r(1)
and S
0
(2) = (1 +i (1))(1 +i(2)) = e
r(1)+r(2)
. We
assume, in addition, that the future interest can take two different values
only. (For a treatment of such models, see, for example, Jarrow (1996).) An
illustration of this model can be found in Figure 3.8, where the interest for
the first period is taken to be i (1) =5% and where the interest for the second
period is either i(2) =4% or i(2) =6%. Let
¡ =P(i(2) =0.04) =1−P(i(2) =0.06).
and assume that 0 -¡ -1, i.e. there is a positive possibility for both events.
The example ¡ =1¡2 is the case where both events are equally likely.
One can now ask questions such as: is is possible to find the price of a
zero coupon bond with maturity T = 2 from these assumptions? First, we
recall that P(2. 2) = 1, independently of the value of the interest. Then, we
consider the question of finding the value of the bond at time 1. If interest
has increased from 5% to 6%, we are in the upper part of Figure 3.8, where
it is possible to deposit money on the savings account and receive an interest
of 6% at the end of the period. The usual arbitrage argument now shows that
3.7 Arbitrage-free pricing in discrete time 81
1
(0.05)
1.05
(0.06)
1.05
(0.04)
1.113
1.113
1.092
1.092
Figure 3.8. Model for the interest (numbers in parentheses) and the savings account S
0
.
the value at time 1 of a zero coupon bond with maturity 2 in this situation
must be given by
P(1. 2) =
1
1.06
≈0.943.
If interest goes down to 4%, we obtain
P(1. 2) =
1
1.04
≈0.962.
However, we still cannot say precisely what the price P(0. 2) at time 0
for the zero coupon bond with maturity 2 should be from the condition of
no arbitrage alone. Actually, there are many different prices for the zero
coupon bond which are consistent with the condition of no arbitrage. These
observations are collected together in Figure 3.9.
In the following we show that the condition of no arbitrage only yields
that
0.898 ≈
0.943
1.05
-P(0. 2) -
0.962
1.05
≈0.916.
However, when the price at time 0 of the zero coupon bond is given, we are
able to derive the price of many other contracts. For example, assume that we
know already at time 0 that we want to invest one unit at time 1. Is it then
possible to ensure that the return does not fall below 4.5%, when the interest
in the lower part of Figure 3.9 is 4%? We will introduce some additional
concepts before we come back to this question.
82 Interest rate theory in insurance
?
(0.05)
0.943
(0.06)
0.962
(0.04)
1
1
1
1
Figure 3.9. Interest (numbers in parentheses) and the price for the zero coupon
bond with maturity 2.
3.7.3 Investment strategies and the value process
An investment strategy (or simply a strategy) is a two-dimensional process
h = (h
0
. h
1
), where h
1
is predictable (i.e. h
1
(t) is known/chosen at time
t −1) and h
0
is adapted (i.e. h
0
(t) is known/chosen at time t). This construc-
tion is illustrated in Figure 3.10, which describes when the number h
1
of
zero coupon bonds purchased and the deposit h
0
on the savings account are
chosen.
The quantity h
1
(t) represents the total number of units of asset number 1
which are in the portfolio at time t −1. In particular, these assets have been
a part of the investment portfolio from time t −1 to time t. The condition
concerning predictability of h
1
ensures that the number of assets held from
time t −1 to time t is fixed at t −1 based on the information available at
this time. In contrast, the deposit h
0
on the savings account is only required
to be adapted, such that the deposit at time t can be decided based on
the additional information which arises from time t −1 to time t. The pair
h(t) =(h
0
(t). h
1
(t)) is also called the portfolio at time t.
At time t −1, the insurance company’s portfolio is given by
h(t −1) =(h
0
(t −1). h
1
(t −1)).
i.e. h
1
(t −1) zero coupon bonds and a deposit on the savings account with
value h
0
(t −1)S
0
(t −1). The discounted value of the portfolio h(t −1) at
3.7 Arbitrage-free pricing in discrete time 83
savings account: h
0
(0) h
0
(1) h
0
(t − 1) h
0
(t) h
0
(T)
bonds: h
1
(1) h
1
(2) h
1
(t) h
1
(t + 1) ·
time: 0 t − 1 t T 1
Figure 3.10. The investment strategy h =(h
1
. h
0
) and times for changes of investments.
time t −1, where we use the savings account S
0
as the discount factor, is
defined by
V(t −1. h) =S
0
(t −1)
−1

h
1
(t −1)S
1
(t −1) +h
0
(t −1)S
0
(t −1)

=h
1
(t −1)X(t −1) +h
0
(t −1).
The process (V(t. h))
t∈]0.1.... .T]
is also called the (discounted) value process
of h.
We now turn to an analysis of the flow of capital, which takes place during
the time interval (t −1. t]; see Figure 3.10. We also refer to the insurance
company as the hedger.
3.7.4 The cost process
Is is essential to describe changes in the value process and to keep track of
whether changes are due to returns on investments or whether new capital
has been added. For this purpose, we introduce the so-called cost process
suggested by Föllmer and Sondermann (1986) and Föllmer and Schweizer
(1988); see also Föllmer and Schied (2002). This process can be applied for
the construction of risk-minimizing strategies; for applications in insurance,
see Chapter 5 or Møller (2002) and references therein.
Consider again the interval (t −1. t]. Immediately after time t −1, the
portfolio h(t −1) is adjusted so that the hedger now holds h
1
(t) bonds. This
is achieved by buying an additional h
1
(t) −h
1
(t −1) bonds, and this gives
rise to the discounted costs
(h
1
(t) −h
1
(t −1))X(t −1).
The new portfolio (h
0
(t). h
1
(t −1)) is held until time t, when new prices
(S
0
(t). S
1
(t)) are announced, and thus the hedger receives the following
discounted gains:
h
1
(t)(X(t) −X(t −1)).
84 Interest rate theory in insurance
Finally, the hedger may at time t decide to change the deposit on the savings
account from h
0
(t −1)S
0
(t) to h
0
(t)S
0
(t) based on the additional information
available at time t. This change leads to the additional discounted costs
h
0
(t)−h
0
(t −1). Thus, we have seen that the change in value of the investment
portfolio can be written as follows:
V(t. h) −V(t −1. h) = (h
1
(t) −h
1
(t −1))X(t −1)
+h
1
(t)(X(t) −X(t −1)) +(h
0
(t) −h
0
(t −1)).
(3.48)
The first and the last terms on the right hand side of Equation (3.48) represent
costs to the hedger, whereas the second term is trading gains obtained from
the strategy h during (t −1. t]. We now introduce the cost process (or the
accumulated cost process) of the strategy h, given by
C(t. h) =V(t. h) −
t
¸
s=1
h
1
(s)AX(s). (3.49)
where we have used the notation AX(s) =X(s) −X(s −1). The cost process
is simply defined as the value of the strategy reduced by trading gains; in
particular, the cost process C(h) satisfies the following relation:
V(t. h) =V(t −1. h) +h
1
(t)(X(t) −X(t −1)) +(C(t. h) −C(t −1. h)).
which corresponds to Equation (3.48). It decomposes changes in the value
process into trading gains and changes in the accumulated cost process, i.e.
additional investments made during |0. t]. We note that C(0. h) = V(0. h),
which says that the initial costs are exactly equal to the amount invested at
time 0.
3.7.5 Self-financing strategies and arbitrage
A strategy is said to be self-financing if the change in the value process given
by Equation (3.48) is generated by trading gains only, i.e. if the portfolio is
not affected by any in- or outflow of capital during the period considered. This
means that any changes in the portfolio have to be made in a cost-neutral
way, in the sense that the purchase of additional bonds must be financed
by reducing the deposit on the savings account by a similar amount. This
condition amounts to requiring that
V(t. h) =V(0. h) +
t
¸
s=1
h
1
(s)AX(s). (3.50)
3.7 Arbitrage-free pricing in discrete time 85
for all t. By inserting Equation (3.50) into Equation (3.49), we see that this
corresponds to saying that the cost process is constant and equal to V(0. h), i.e.
equal to the value of the initial investment made at time 0. This characterizes
the self-financing strategies in terms of the cost process.
An arbitrage is a self-financing strategy h such that
V(0. h) =0. P(V(T. h) ≥0) =1 and P(V(T. h) >0) >0. (3.51)
The interpretation is exactly the same as in Example 3.11; it corresponds to
receiving a free lottery ticket. With this strategy, one can invest zero units
at time 0 and receive at time T the non-negative amount V(T. h), which is
strictly positive with positive probability.
3.7.6 Hedging and attainability
The insurer is assumed to invest on the financial market in order to control
the risk associated with some liability H; i.e., the insurer is hedging against
some risk in order to control or eliminate the risk. In some situations (within
certain models) it is possible to determine a self-financing strategy which
generates or replicates the liability completely. This is the case if there exists
a self-financing strategy h which sets out with some amount V(0. h) and has
terminal value V(T. h) = H at time T. In this case, the initial value V(0. h)
is the only reasonable price for the liability H; such claims are also said to
be attainable, and V(0. h) is called the arbitrage-free price of H. This is the
case in the example considered in Section 3.7.8. However, in many cases the
hedger’s liabilities cannot be hedged perfectly using a self-financing strategy,
and this leaves open the question of how to choose an optimal trading strategy.
For a theoretical discussion on the choice of trading strategy for insurance
contracts with financial risk, see Møller (2001a, 2002).
3.7.7 Equivalent martingale measures and absence of arbitrage
Typically we would like to verify that a given model does not allow for
arbitrage possibilities. Here, the concept of equivalent martingale measures
plays a central role.
Equivalent martingale measures
When we specify how the price processes S
0
and S
1
develop, we also have
to associate probabilities to the event that prices attain any given value. This
fixes the distribution of the future prices and leads to a probability measure P.
86 Interest rate theory in insurance
More generally, we specify the probabilities for a certain class of events
A (for example, that tossing a coin leads to “head” as in Section 3.2.1). We
write P(A) for the “probability of A under P,” where we have added “under
P” in order to underline the specific probability and the fact that one could
as well speak of the probability of A under another measure Q, say. This
situation can, for example, be compared with the way the actuary handles
simultaneously various bases for the mortality of a policy holder. Actually, it
is possible to formalize this change of mortality intensity by exactly the same
method as the one sketched below. For more details, see Section A.1 of the
Appendix.
It is essential to introduce another probability measure Q, which associates
to events probabilities which might differ from the ones under P. At the
moment, we only require that the two probability measures are equivalent.
Loosely speaking, this means that the measures agree on whether an event
is totally unlikely or not, which means that if P(A) = 0, we also have that
Q(A) =0 (and vice versa). If we again compare with the situation with two
different mortality intensities, then it means that the mortality intensity is
µ
P
if we use the probability measure P and µ
Q
if we use the probability
measure Q. Basically, the two measures P and Q are now equivalent if
we have that µ
P
is zero, if and only if µ
Q
is zero. In this part, we deal
primarily with the development of bond prices over time, and the two different
probability measures represent different expectations for the changes in bond
prices for a given period. Before we explain why the concept “equivalent
martingale measures” appears in the title of this section, we briefly recall
what a martingale is.
What is a martingale?
Let us again consider the discounted price process X, which is the ratio
between the price of a bond S
1
and the value S
0
of one unit invested at time
0 of the savings account. We can now ask questions such as: what is the
expected (discounted) price of the bond at time u? Here, one has to be a little
more precise and specify the probability measure that one wishes to use for
the calculations – P or Q? If we work with P, the answer is E
P
|X(u)], and
if we use Q the answer is E
Q
|X(u)]. We can modify the question slightly
and ask: what is the expected discounted price under Q for the bond at time
u, given that we take into consideration the amount (t) of information
available at time t -u? This quantity is
E
Q
|X(u)(t)]. (3.52)
3.7 Arbitrage-free pricing in discrete time 87
the conditional expected value under Qof X(u) given the information available
at t. The process X is now said to be a martingale (under Q), if Equation (3.52)
is exactly equal to the discounted price at time t, X(t), i.e.
E
Q
|X(u)(t)] =X(t). (3.53)
for all t ≤ u. This condition becomes slightly more transparent in the case
where X is a Markov process, since Equation (3.53) then reduces to
E
Q
|X(u)X(t)] =X(t). (3.54)
We interpret this as follows. If X(t) and X(u) represent the discounted price
at time t (today) and time u (tomorrow) of a bond, and if Q is a martin-
gale measure, the expected discounted price tomorrow (calculated under the
probability measure Q) is identical to the price today.
Equivalent martingale measures
An equivalent probability measure Q is said to be an equivalent martingale
measure if Xis a martingale under Q, i.e. if Equation (3.53) is satisfied for all t -
u. We emphasize that the discounted bond price X is typically not a martingale
under the true probability measure P, so that P normally cannot be applied as an
equivalent martingale measure. However, the measure P is important since, in
particular, it determines the events that are completely unlikely, i.e. if an event
has probability zero. The importance of the concept of equivalent martingale
measures is illustrated below. Here we introduce the following assumption.
Assumption 3.12 There exists at least one equivalent martingale measure.
Martingale measures ensure absence of arbitrage
It is actually not so difficult to show that the existence of an equivalent mar-
tingale measure is sufficient to exclude the possibility of arbitrage possibilities
in the model, i.e. that there cannot be strategies which satisfy Equation (3.51).
Consider some equivalent martingale measure Q, such that the discounted
price process X is a Q-martingale; i.e., for any t -u ∈ ]0. 1. . . . . T], we have
that E
Q
|X(u)(t)] = X(t). We want to show that there cannot exist any
self-financing strategies h with V(0. h) =0 and where
V(T. h) =V(0. h) +
T
¸
s=1
h
1
(s)AX(s) (3.55)
88 Interest rate theory in insurance
is non-negative and strictly positive with probability 1. To see that such
strategies cannot exist, it is sufficient to show that
E
Q
|V(T. h)] =V(0. h). (3.56)
Why does this exclude arbitrage possibilities? Well, if V(T. h) ≥0 and
E
Q
|V(T. h)] =V(0. h) =0.
we must have that Q(V(T. h) =0) =1, and hence P(V(T. h) =0) =1, since
P and Q are equivalent.
We still need to show that Equation (3.56) is satisfied for any self-financing
strategy h. Here, we need the special construction of the investment strategy
h, which ensures that h
1
is predictable, i.e. that the number h
1
(s) of bonds
in the portfolio during the period (s −1. s] is known at time s −1. Using this
property, we obtain the following:
E
Q
¸
h
1
(s) AX(s)

(s −1)
¸
=h
1
(s) E
Q
| AX(s) (s −1)] =0.
where we have moved h
1
(s) outside the expected value. (Mathematically, we
have used that h
1
(s) is (s −1)-measurable.) The second equality follows
by noting that X is a Q-martingale, so that
E
Q
|AX(s)(s −1)] =E
Q
|X(s) −X(s −1)(s −1)]
=X(s −1) −X(s −1) =0.
In fact, this shows that the value process given in Equation (3.50) for a
self-financing strategy is also a Q-martingale, i.e.
E
Q
|V(u. h)(t)] =V(t. h).
for all t ≤u. If we insert u =T and t =0, we see that
E
Q
|V(T. h)] =V(0. h).
Thus, under Assumption 3.12 there exist no strategies h that satisfy
Equation (3.51).
Pricing with an equivalent martingale measure
We now consider some liability which specifies the discounted payoff H,
which can be replicated by using a self-financing strategy h, i.e.
H =V(T. h) =V(0. h) +
T
¸
t=1
h
1
(t) AX(t). (3.57)
3.7 Arbitrage-free pricing in discrete time 89
In Section 3.7.6 we introduced the notion of an arbitrage-free price and we
explained that this price is equal to V(0. h), the price needed in order to
generate H with the self-financing strategy. We can now use the martingale
measure Q to compute expected values in Equation (3.57). Since the value
process V(h) is a martingale under Q, we see that
E
Q
|H] =E
Q
|V(T. h)] =V(0. h). (3.58)
This very important result shows us that the arbitrage-free price for the liability
can actually be calculated as the expected value of the discounted payments,
computed under the equivalent martingale measure Q.
3.7.8 The two-period example revisited
Let us return to the example considered in Section 3.7.2 and assume that
we have observed the price P(0. 2) at time 0 of the zero coupon bond with
maturity 2. The other prices at times 1 and 2 have already been established; see
Section 3.7.8. Do there exist martingale measure(s)? How can we determine
a martingale measure?
Define a probability measure Q via
q =Q(i(2) =0.04) =1−Q(i(2) =0.06).
where 0 ≤q ≤1. One can now start to think about whether Q is equivalent to
the original measure P, i.e. if the two measures agree about whether an event
has probability 0 or a strictly positive probability. Under P, the probability of
i(2) attaining the values 0.04 and 0.06 is strictly positive. Hence this should
also be the case under Q; this observation gives the condition 0 -q -1.
Next, we check if it is possible to choose q in such a way that Q actually
becomes a martingale measure. Here, we need to verify that the process
X(t) =
P(t. 2)
S
0
(t)
is indeed a martingale under Q. Thus, it should be shown that
E
Q
¸
P(u. 2)
S
0
(u)

(t)
¸
=
P(t. 2)
S
0
(t)
. (3.59)
for 0 ≤t -u ≤2. First, consider the case t =1 and u =2, so that P(u. 2) =1
and S
0
(u) =S
0
(t) (1+i(2)). Since the interest i(2) is assumed to be known
at time t =1, Equation (3.59) implies that
1
1+i(2)
=P(1. 2).
90 Interest rate theory in insurance
which is exactly the formula that we derived already in Section 3.7.2. For
u =1 and t =0, Equation (3.59) states that
E
Q
¸
P(1. 2)
S
0
(1)
¸
=
P(0. 2)
S
0
(0)
=P(0. 2). (3.60)
In principle, we should include here the information (0) available at time 0.
However, in our model there is no (non-trivial) information available at time
0, so this can be neglected. In order to be able to calculate the expected value
on the left hand side of Equation (3.60), we note that the probability (under
Q) that i(2) is equal to 0.04 is q (in this case, P(1. 2) =0.962), and that the
probability that i(2) is equal to 0.06 is 1−q (in this case, P(1. 2) =0.943).
Thus we obtain the following condition:
P(0. 2) =E
Q
¸
P(1. 2)
S
0
(1)
¸
=
1
1.05
(q 0.962+(1−q) 0.943) .
which has the unique solution
q =
1.05P(0. 2) −0.943
0.962−0.943
.
In particular, this shows that if 0.943 -1.05P(0. 2) -0.962, then 0 -q -1
and hence Q is indeed an equivalent martingale measure in this case. We see
moreover that the martingale measure in this example is uniquely determined
from the prices which are given on the market. We can say that the market
determines the martingale measure uniquely in this case.
A very simple interest rate guarantee
Consider now the example where we are interested in a protection against
the situation where the interest in the second period falls below 4.5%, say.
Assume more precisely that we know that we will invest one unit at time
1 (for example because we receive some insurance premiums at this time),
and that we want a guarantee which helps us in the scenario where i(2) is
4%. We therefore consider the contract which pays at time 2 the amount
max(1.045; 1+i(2)).
The discounted payment is given by
H =
max(1.045; 1+i(2))
S
0
(2)
=
1
1.05(1+i(2))
(1+i(2) +(0.045−i(2))
+
).
(3.61)
3.7 Arbitrage-free pricing in discrete time 91
where ()
+
is the positive part. According to Equation (3.58), the arbitrage-free
price at time 0 for this contract is as follows:
E
Q
|H] =
1
1.05
+
1
1.05

q
0.005
1.04
+(1−q)0

=
1
1.05
+q
0.005
1.04· 1.05
.
We see that the price consists of two terms, where the second term is the
price for the guarantee. This price depends on the price of the zero coupon
bond; see Figure 3.11.
We end this example by showing how the interest rate risk can alternatively
be hedged by buying zero coupon bonds and by using the savings account.
More precisely, we construct a self-financing strategy h with a terminal
value V(T. h), which coincides with the guarantee part of Equation (3.61);
see Equation (3.57). Since the interest rate i(2) is known at time 1, we can
construct a figure similar to Figures 3.7 and 3.8 with prices for the interest
rate guarantee. This is illustrated in Figure 3.12.
A self-financing strategy which generates the guarantee part of the contract,
Equation (3.61), must satisfy the following equation:
V(0. h) +h
1
(1) AX(1) +h
1
(2) AX(2) =
1
1.05(1+i(2))
(0.045−i(2))
+
.
In our simple example, AX(2) =0 and
AX(1) =
P(1. 2)
1.05
−P(0. 2).
0.900 0.905 0.910 0.915
0.000
0.001
0.002
0.003
0.004
zero coupon bond price
p
r
i
c
e

f
o
r

i
n
t
e
r
e
s
t

r
a
t
e

g
u
a
r
a
n
t
e
e
Figure 3.11. The price q (0.005¡1.04· 1.05) for the interest guarantee as a
function of the price at time 0 of the zero coupon bond with maturity 2.
92 Interest rate theory in insurance
q
0.005
1.04·1.05
(0.05)
0
(0.06)
0.005
1.04
(0.04)
0
0
0.005
0.005
Figure 3.12. Model for the interest i(t) (numbers in parentheses) and the price
for the interest rate guarantee.
so that we need to determine h
1
(1) and h
0
(0) such that
V(0. h) +h
1
(1)

P(1. 2)
1.05
−P(0. 2)

=
1
1.05(1+i(2))
(0.045−i(2))
+
.
This leads to two equations (one for each of the possible outcomes for the
future interest):
h
0
(0) +h
1
(1)P(0. 2) +h
1
(1)

0.962
1.05
−P(0. 2)

=
0.005
1.05· 1.04
;
h
0
(0) +h
1
(1)P(0. 2) +h
1
(1)

0.943
1.05
−P(0. 2)

=0.
These two equations are solved as follows:
h
1
(1) =
0.005
1.04(0.962−0.943)
≈0.253;
h
0
(0) =−
0.005· 0.943
1.05· 1.04(0.962−0.943)
≈−0.227.
The example shows that it is indeed possible to hedge the interest guarantee
by buying 0.253 units of the zero coupon bond and by borrowing 0.227 units
from the savings account. With this strategy, the insurer is protected against
the drop in the interest rate. We point out that the price for the interest rate
guarantee, see Figure 3.12, and the investment strategy are both independent
3.8 Models for spot rate in continuous time 93
of the original probability measure P. The measure P is only used to specify
whether an event is completely unlikely (has probability 0) or not.
For a more detailed study of discrete-time bond markets, we refer to Jarrow
(1996).
3.8 Models for the spot rate in continuous time
In this section we briefly describe some so-called diffusion models for the
spot rate. We assume that the change in spot rate during a small time interval
(t. t +At] can be approximated by
r(t +At) −r(t) =Ar(t) ≈ì(t. r(t))At +u(t. r(t))AW(t). (3.62)
Here, we have used the notation AW(t) =W(t +At) −W(t); ì are u known
functions. We assume that the process W is a Brownian motion, which means
that W has independent, normally distributed increments, with W(t)−W(s) ∼
N(0. t −s) (mean 0 and variance t −s), and that W is continuous. In particular,
this means that AW(t) ∼ N(0. At).
It is possible to simulate r by choosing some interval length At, simulating
independent normally distributed random variables AW(t), and inserting these
into Equation (3.62) for the computation of the increments for r.
To underline that Equation (3.62) is only valid for very small (infinitesimal)
time intervals, we also use the following notation:
dr(t) =ì(t. r(t)) dt +u(t. r(t)) dW(t). (3.63)
Similarly, we can integrate Equation (3.63) and write r in the following form:
r(t) =r(0) +

t
0
ì(s. r(s)) ds +

t
0
u(s. r(s)) dW(s). (3.64)
The savings account S
0
is defined by letting S
0
(0) =1 and
dS
0
(t) =r(t) S
0
(t) dt.
It is well known that S
0
is then given by
S
0
(t) =exp

t
0
r(t) dt

.
As mentioned above, (t) is the information available at time t. We take
(t) =u]W(u). u ≤t].
which means that we observe the Brownian motion W. Typically this corre-
sponds to observing the interest r.
94 Interest rate theory in insurance
In Section 3.7 we showed that prices of zero coupon bonds and derivatives
could be calculated as an expected value under an equivalent martingale mea-
sure Q. In particular, prices are not determined by the probability measure P.
In addition, we mentioned that the measure Q should be determined from the
zero coupon bond prices that are given on the market. We now consider some
martingale measure Q and assume that
dr(t) =µ(t. r(t)) dt +u(t. r(t)) dW(t). (3.65)
where the function ì has been replaced by another function µ, and where W is
a Brownian motion under Q. The function u appears in both Equations (3.63)
and (3.65). We refer to Equation (3.65) as the “dynamics for r under Q.”
With this martingale measure Q, the price at time t of a zero coupon bond
maturing at time T is given by
P(t. T) =E
Q
¸
S
0
(t)
S
0
(T)

(t)
¸
=E
Q
¸
exp

T
t
r(t) dt

(t)
¸
; (3.66)
see, for example, Equation (3.59) for the similar result in discrete time.
This ensures that the discounted price process P(t. T)¡S
0
(t) is indeed a
Q-martingale.
Estimation of parameters
The functions ì and u, which describe the behavior of the spot rate under the
probability measure P, are typically specified in such a way that they depend
on some parameters. These parameters can be estimated by observing the
interest rate over a certain period and then finding the value which describes
the observations as well as possible, for example via maximum likelihood
estimation. In contrast, the function µ describes how the interest rate evolves
under the martingale measure Q, and this function cannot be estimated by
observing the interest rate alone. Here one needs to include some observed
zero coupon bond prices, compute Equation (3.66) for various choices of
parameters, and then find the parameters which give the best description of
the observed prices.
Some classical examples of interest rate models of the form in
Equation (3.65) are the Vasiˇ cek model, Equation (3.67) (see Vasiˇ cek (1977)),
3.8 Models for spot rate in continuous time 95
and the Cox–Ingersoll–Ross model, Equation (3.68) (see Cox, Ingersoll and
Ross (1985)):
dr(t) =(l −or(t))dt +u dW(t). (3.67)
dr(t) =o(l −r(t)) dt +u

r(t) dW(t). (3.68)
Both models have the special property that the interest rate returns to a certain
level. For the Vasiˇ cek model with parameterization, Equation (3.67), this level
is given by l¡o, and for the Cox–Ingersoll–Ross model, Equation (3.68), this
level is l. These two models are examples of so-called affine models, which
allow for relatively simple formulas for zero coupon bond prices.
3.8.1 Affine models
What is an affine model? How can prices for zero coupon bonds be calculated
in the Vasiˇ cek and Cox–Ingersoll–Ross models? Is it possible to determine
the instantaneous forward rates in these models?
An interest rate model of the form given in Equation (3.65) is said to be
affine if
µ(t. r(t)) =o(t)r(t) +8(t).
u(t. r(t)) =

~(t)r(t) +o(t).
where o. 8. ~ and o are known functions. (Thus, for fixed t, µ(t. ·) and
u
2
(t. ·) are affine functions of r.)
We have the following general result for the price of a zero coupon bond
with maturity T (see, for example, Björk (2004, Proposition 17.2) for a proof
of this result):
P(t. T) =E
Q
¸
exp

T
t
r(t) dt

(t)
¸
=exp(A(t. T) −B(t. T)r(t)).
(3.69)
where A and B solve the following equations:
o
ot
B(t. T) +o(t)B(t. T) −
1
2
~(t)(B(t. T))
2
=−1.
o
ot
A(t. T) =8(t)B(t. T) −
1
2
o(t)(B(t. T))
2
.
with B(T. T) =A(T. T) =0. The equations can be solved by first determining
the function B from the first equation and then inserting this solution in the
second equation in order to find A.
96 Interest rate theory in insurance
The instantaneous forward rates at time t can be found via Equation (3.29)
from Definition 3.5:
1(t. T ) =−
o
oT
log(P(t. T )) =r(t )
o
oT
B(t. T ) −
o
oT
A(t. T). (3.70)
The Vasiˇ cek model
With the model given in Equation (3.67), we obtain by solving the differential
equations for A and B the following:
B(t. T ) =
1
o

1−e
−o(T−t)

.
A(t. T) =
(B(t. T ) −T +t )(ol −(1¡2)u
2
)
o
2

u
2
(B(t. T ))
2
4o
.
The forward rates at time t can be determined by differentiating these expres-
sions with respect to T and then inserting the results into Equation (3.70). In
this way, it can be shown that
1(t. T ) = r(t )e
−o(T−t)
+

l
o

u
2
2o
2

1−e
−o(T−t)

+ e
−o(T−t)
2u
2
B(t. T )
4o
. (3.71)
Since e
−o(T−t)
→0 as T →, we see that B(t. T ) →1¡o. This shows that
the forward rates given by Equation (3.71) converge to
l
o

u
2
2o
2
. (3.72)
as the maturity goes to infinity. Figure 3.13 presents a few examples of the
shape of the forward rate curve. We see that the shape depends crucially on
the level of the interest rate r(t) at the time considered, in particular, whether
the interest rate r(t) is above or below the level given by Equation (3.72).
A couple of simulations for r can be found in Figure 3.14. The Vasiˇ cek
model has the rather unfortunate property that it allows for negative interest
rates. Similarly, the forward rates might become negative, a phenomenon
which is most likely for large values of u. This is not the case for the
Cox–Ingersoll–Ross model, which does not allow for negative interest rates.
3.8 Models for spot rate in continuous time 97
0 5 10 15 20 25 30
0.00
0.02
0.04
0.06
0.08
maturity
(a)
(b)
f
o
r
w
a
r
d

r
a
t
e
0 5 10 15 20 25 30
0.00
0.02
0.04
0.06
0.08
maturity
f
o
r
w
a
r
d

r
a
t
e
Figure 3.13. Forward rates in the Vasiˇ cek model with high initial interest
r(0) = 0.07 (solid curves) and low initial interest r(0) = 0.03 (dashed curves).
Parameters used: o = 0.36, l¡o = 0.06. (a) Low standard deviation, u = 0.03;
(b) high standard deviation, u =0.09.
10 5 0 15 20 25 30
0.00
0.05
0.10
0.15
time
s
p
o
t

r
a
t
e
Figure 3.14. Two possible simulations for r in the Vasiˇ cek model with initial
interest rate r(0) =0.03. Parameters: o =0.36, l¡o =0.06, u =0.03.
98 Interest rate theory in insurance
The Cox–Ingersoll–Ross model
With the model given in Equation (3.68), we obtain
B(t. T) =
2(e
£(T−t)
−1)
(£ +o)(e
£(T−t)
−1) +2£
.
A(t. T) =
2ol
u
2
log

2£e
(o+£)
T−t
2
(£ +o)(e
£(T−t)
−1) +2£

.
where
£ =

o
2
+2u
2
.
Again, it is possible to determine the forward rates by using Equation (3.70).
However, the formulas turn out to be considerably more complicated than the
ones obtained for the Vasiˇ cek model.
3.9 Market values in insurance revisited
We end this chapter by illustrating how the market values of the guaran-
teed payments may be calculated as the expected value under an equivalent
martingale measure of the discounted payments. In Sections 3.2 and 3.3 we
presented methods for deriving the market value of the guaranteed payments
in a model where the interest rate could be stochastic. These results were
stated in terms of either prices of zero coupon bonds or forward rates. In Sec-
tions 3.7 and 3.8 we showed how the price at time t of a zero coupon bond
with maturity n could be written in the following form:
P(t. n) =E
Q
¸
S
0
(t)
S
0
(n)

(t)
¸
.
where the expected value is calculated by using some equivalent martingale
measure Q. This result can now be inserted into the formulas derived in the
main example in Sections 3.2.7 and 3.3.5. The market value at time u of the
payments guaranteed at time t can hence be rewritten as follows:
V
g
(t. u) =l
a
(t) E
Q
¸
S
0
(u)
S
0
(n)

(u)
¸
n−u
¡
x+u
+

n
u
E
Q
¸
S
0
(u)
S
0
(s)

(u)
¸
s−u
¡
x+u
(µ(x+s)l
ad
−r)ds. (3.73)
3.9 Market values in insurance revisited 99
which can be recast as
V
g
(t. u) =E
Q
¸
l
a
(t)
S
0
(u)
S
0
(n)
n−u
¡
x+u
+

n
u
S
0
(u)
S
0
(s)
s−u
¡
x+u
(µ(x+s)l
ad
−r)ds

(u)
¸
. (3.74)
In models with a continuously added interest rate, we can use the fact that
S
0
(u)
S
0
(s)
=exp

s
u
r(t) dt

to rewrite Equation (3.74) as follows:
V
g
(t. u) =E
Q
¸
l
a
(t) exp

n
u
r(t) dt

n−u
¡
x+u
+

n
u
exp

s
u
r(t) dt

s−u
¡
x+u
(µ(x+s)l
ad
−r) ds

(u)
¸
.
(3.75)
This brings to the surface the consequences for the market value, Equa-
tion (3.24), of allowing for stochastic interest rates. In these situations, mar-
ket values are calculated as expected values under the equivalent martingale
measure.
The only randomness in Equation (3.75) is that related to the develop-
ment of the interest rate, i.e. the uncertainty related to the policy holder’s
lifetime has been averaged out. However, it is possible to work with the pay-
ment process B(t. s) introduced in Chapter 2, which described the payments
guaranteed at time t. This process is defined by
dB(t. s) =−rI(s) ds +l
ad
dN(s) +l
a
(t) I(s) de(s. n).
where I(s) is the indicator for the event that the insured is alive at time s, N
is a counting process which counts the number of deaths (and hence which
is either zero or one) and e(s. n) = 1 if s ≥ n and zero otherwise. With this
notation, Equation (3.75) can finally be rewritten as follows:
V
g
(t. u) =E
Q
¸

n
u
exp

s
u
r(t) dt

dB(t. s)

(u)
¸
. (3.76)
Here, the uncertainty related to the policy holder’s lifetime and the uncertainty
related to the future development of the interest rate appear simultaneously.
100 Interest rate theory in insurance
The market value is calculated directly as an expected value under a martingale
measure of the random variable

n
u
exp

s
u
r(t) dt

dB(t. s). (3.77)
From a theoretical point of view, Equation (3.77) is the natural starting point
for a discussion of market values in life and pension insurance. This analysis
is continued in Chapter 6.
This chapter has focused on determining the market value for the guar-
anteed payments in situations where the interest rate is stochastic. In this
situation, it is natural to replace the usual discount factors by zero coupon
bond prices. Moreover, we have shown how forward rates can be applied
to derive market values and the individual bonus potential. Finally, it has
been demonstrated that the market values can alternatively be computed by
using equivalent martingale measures. The next chapter investigates how the
market value of the total liability is affected by changes in the interest and
in other assets, for example stocks. Such considerations typically rely on a
specification of the bonus and investment policy of the insurance company.
4
Bonus, binomial and Black–Scholes
4.1 Introduction
In Chapter 2 we discussed some aspects of valuation assuming only one pos-
sible investment with a deterministic interest rate. In Chapter 3 we introduced
a stochastic interest rate and a bond market and we discussed the conse-
quences for valuation in general and for valuation of guaranteed payments in
particular. In this chapter we again assume a deterministic interest rate, but,
in return, we introduce the possibility of investing in stocks and study the
total reserve including the reserve for guaranteed payments. In Section 4.6
we comment on the combination of stochastic interest rates and investment
in stocks.
The total reserve in connection with a life insurance contract can, under
certain conditions, be calculated using a simple retrospective accumulation.
The condition is that the total reserve which has been accumulated at the
termination of the contract equals the pension sum paid out. We consider the
type of insurance where the surplus is accumulated in the technical reserve
leading to an increasing pension sum. Here, the condition is that the undis-
tributed reserve, which is the total reserve minus the technical reserve, at the
termination of the contract equals zero. The condition and its consequences
are formalized and studied in Chapter 2.
One very simple situation in Chapter 2 was the financial market, which
consists of one investment possibility only, namely the possibility of invest-
ing in the risk-free interest rate. Furthermore, this risk-free interest rate is
assumed to be deterministic. In Chapter 3 we introduced a more realistic
stochastic model for the interest rate and we discussed the consequences for
the entries in the market balance scheme, in particular for the market reserve
for guaranteed payments and the individual bonus potential. In this chapter,
we introduce investment possibilities in so-called risky assets or stocks and
101
102 Bonus, binomial and Black–Scholes
study their consequences for the total reserve. The collective bonus potential
is the difference between the total reserve and the larges of either the technical
reserve or the market reserve for guaranteed payments. Thus, the collective
bonus potential is indirectly determined by calculating the total reserve.
The purpose of this chapter is to develop theoretically substantiated meth-
ods for the valuation of the bonus obligations under investment in stocks. The
starting point is two classical stock price models in mathematical finance:
the binomial model and the Black–Scholes model. The binomial model is
the simplest non-trivial stock model one can imagine; pricing in the bino-
mial model is not much more than the solution of two equations with two
unknowns. This provides a simple framework for the discussion of the various
difficulties arising from a market valuation of insurance liabilities.
The Black–Scholes model is a classical stock price model based on nor-
mally distributed returns over small time intervals. The concepts of financial
mathematics are mathematically much more involved in the Black–Scholes
model than in the binomial model. The power of the Black–Scholes model is
that, in spite of its complexity, it has some simple and appealing qualities that
provide unique answers to certain pricing problems. This makes it a popular
stock price model in practice. The authors of the Black–Scholes model were
awarded the 1997 Nobel Prize in Economics.
The Black–Scholes model has been the preferred model for the financial
market in literature connecting bonus in life insurance with financial the-
ory. The starting point is typically a Black–Scholes market, within which
the authors give a formalization of the special payment streams appearing in
life insurance contracts with bonus and possibly other options such as the
surrender option. Seminal references are Briys and de Varenne (1994, 1997),
which describe the bonus option as a financial derivative within the termi-
nology of financial mathematics. Further important contributions are Grosen
and Jørgensen (1997, 2000, 2002), Hansen and Miltersen (2002), Miltersen
and Persson (2000, 2003), and also Bacinello (2001) should be mentioned.
A complete reference list should also contain Møller (2000) and Steffensen
(2001), which deal with the issues on a broader level.
What mainly distinguishes the abovementioned contributions from each
other is the way in which the bonus option is formalized, or in another way:
precisely how is the value of assets reflected in the bonus that is paid out?
This is a fundamental question in this chapter. Furthermore, we go through
some calculations for certain relevant formalizations.
As in Chapter 2, we consider an endowment insurance paid by a level
premium, in which dividends are used to increase the pension sum. However,
we disregard the mortality risk by setting the mortality intensity to zero.
4.2 Discrete-time insurance model 103
We wish to deal with the financial aspects exclusively. With appropriate
arguments and an assumption of diversified mortality risk, all formulas can
be adapted to a situation including mortality. The actuarial notation is limited
to a discount factor v and a t-year annuity o
t
, which both appear with
decorations depending on the underlying interest rate. The same notation is
used in discrete and continuous time; the relevant version will be obvious
from the context.
In Section 4.2 we describe the relevant quantities in the market balance
scheme in a discrete- time model. We calculate the quantities for one for-
malization of the connection between the bonus and the stock market. In
Section 4.3 we describe the binomial model in general. In Section 4.4 the
Black–Scholes model is described and the celebrated Black–Scholes equa-
tion and Black–Scholes formula are stated and partly derived. In Section 4.5
we show how the Black–Scholes model, the Black–Scholes equation and
the Black–Scholes formula may play important roles in the valuation of life
insurance contracts with bonus. In Section 4.6 we discuss briefly some gen-
eralizations of the considered market models.
The text contains a number of repetitions of definitions and arguments.
This emphasizes the cross-sectional similarities. The notions of discrete- and
continuous-time financial mathematics contain similarities. Nevertheless, they
are introduced in both connections such that both similarities and differences
stand out. Correspondingly, the discrete- and continuous-time insurance mod-
els are closely connected.
4.2 Discrete-time insurance model
In this section we consider a discrete-time insurance model and the dynamics
of various quantities of relevance for valuation within this model. The model
is very simple and not very realistic, but it shows a lot of the problems
that appear when trying to valuate insurance claims by means of so-called
arbitrage arguments. An arbitrage argument is the argument for the value
of a claim that if it had any other value, it would be possible, by tactical
investments, to obtain risk-free gains beyond what is offered by the financial
market.
We formalize the notion of arbitrage and other concepts from mathematical
finance in Section 4.3, where we give a systematic introduction to the binomial
model. However, we start in this section by studying a relatively complicated
example of a product which can be treated within the binomial model. Hereby,
the perspectives of the binomial model are emphasized to readers who prefer
104 Bonus, binomial and Black–Scholes
to think of finance in terms of life insurance and bonus. Møller (2001a) shows
some applications of the binomial model to unit-linked insurance.
The insurance contract that we consider in this section is a modification
of the main example in Chapter 2, such that it appears as a purely financial
product in discrete time and such that investments, on the other hand, take
place in a so- called binomial market introduced below. A lot of the quantities
relate to quantities in Chapter 2, to which we refer for an interpretation. The
insurance sum paid out at time n is paid for by an annuity-due with premium
payment r. We speak of the periods as years, but they could cover other time
intervals.
The technical reserve V

and the second order basis that is made up by the
second order interest rate r
o
are connected by the following equation which
accumulates the technical reserve:
V

(t) =

1+r
o
(t)

V

(t −1) +r1
(t-n)
. (4.1)
V

(0) =r.
We also speak of the second order interest rate as the bonus interest rate and
a process of bonus interest rates

r
o
(t)

t=1.... .n
as a bonus strategy.
The technical reserve and the first order basis which is made up by the
first order interest rate r

are connected by the following equation which
accumulates the technical reserve:
V

(t) =(1+r

) V

(t −1) +r1
(t-n)
+o(t) . (4.2)
V

(0) =r.
where the dividend payment o(t) is given by
o(t) =

r
o
(t) −r

V

(t −1) .
such that the deposits stemming from Equations (4.1) and (4.2) coincide.
The sum l (t) guaranteed at time t is determined for a given technical
reserve V

(t) by the equivalence relation, as follows:
V

(t) =l (t) (v

)
n−t
−ro

n−t
.
In Chapter 2 we introduced the total reserve U, which is obtained by an
accumulation by the real basis:
U (t) =(1+r) U (t −1) +r1
(t-n)
.
U (0) =r.
4.2 Discrete-time insurance model 105
Here, we let r be constant and, in particular, known at time t −1. Thus, at time
t −1, we know which investment gains we have in prospect the following
year. We assume that r ≥r

.
The idea now is to generalize this construction such that we have the
opportunity to invest part of the money in an asset with risk. The most simple
asset with risk is an asset which, given its value at time t −1, at time t takes
one of two values but such that the value at time t is not known at time t −1.
This value is not known until time t. We let S (t) denote the price of the risky
asset at time t and let the dynamics of S be given by
S (t) =(1+Z(t)) S (t −1) .
Here, Z(t) is a stochastic variable which takes the value u (for up) with the
probability ¡(u) and the value J (for down) with the probability ¡(J). We let
the company buy r(t) of these assets at time t −1. Obviously, r(t) must be
decided at time t −1. The company cannot wait until time t, experience the
new asset price and then decide how many assets to buy at time t −1. This is
the intuitive content of the property predictable, which we also require from
our investments later on.
The total reserve U at time t is the value U at time t −1 plus premiums
plus capital gains. The capital gains consist of interest earned on the part
of U which was not invested in the risky asset at time t −1, U (t −1) −
r(t) S (t −1), and the change of the price of the portfolio of risky assets
which follows from changes in asset prices. Thus,
U (t) = U (t −1) +r (U (t −1) −r(t) S (t −1))
+r(t) (S (t) −S (t −1)) +r1
(t-n)
= (1+r) U (t −1) +r(t) S (t −1) (Z(t) −r) +r1
(t-n)
.
U (0) = r.
The undistributed reserve X is given residually by
X(t) =U (t) −V

(t) .
and we obtain the following:
X(t) =(1+r) X(t −1) +r(t) S (t −1) (Z(t) −r) +c (t) −o(t) .
where
c (t) =(r (t) −r

(t)) V

(t −1) .
Thus, the value X at time t is the value of X at time t −1 plus capital gains
plus the surplus contribution c (t) minus the dividend payment o(t).
106 Bonus, binomial and Black–Scholes
In Chapter 2, we also introduced the following prospective quantity for
the total reserve and for the undistributed reserve:
V (t) =l (n) v
n−t
−ro
n−t
.
This reflects the fact that the reserve is set aside to meet future obligations.
The problem with this quantity is that, in general, it requires a formalization
of the future bonus strategy, and we want to avoid that. Very appropriately,
if we arrange the bonus strategy such that X(n) =0, then V (t) =U (t); see
Chapter 2. By fulfilment of the X(n) =0, we can work with the retrospective
quantities exclusively and disregard the future bonus strategy.
The introduction of risky investments makes the situation more difficult.
If the bonus strategy is linked to the gains obtained by risky investments, we
do not in general know l (n) at time t.
On one hand, we wish that l (n), in one way or another, is linked to
the development in asset prices. On the other hand, if we assume that it
is not linked to anything else, in the sense that the development in asset
prices determine l (n) completely, then this can be considered as a so-called
contingent claim, no matter how complex the link may be. Hereafter, the
mathematics of finance concludes a modification of the prospective quantity
as follows:
V (t) =E
Q
t
|l (n)] v
n−t
−ro
n−t
. (4.3)
Here, E
Q
t
denotes the expectation given the development in the financial
market until time t and under a very special probability measure Q, under
which the outcomes that the risky asset goes up and down, respectively,
happen with the following probabilities:
q (u) =
r −J
u−J
.
q (J) =
u−r
u−J
.
(4.4)
For now, we pull this probability measure from a hat, but later we discover
arguments for its construction.
Finally, we need to define the market value of the payments guaranteed at
time t. The following definitions are in correspondence with Equation (4.3).
The market value of the guaranteed payments, V
g
(t), is given by
V
g
(t) =l (t) v
n−t
−ro
n−t
.
4.2 Discrete-time insurance model 107
The market value of the unguaranteed payments, also referred to as the bonus
potential V
b
(t), is given by
V
b
(t) =

E
Q
t
|l (n)] −l (t)

v
n−t
.
In correspondence with the valuation formula, Equation (4.3), and the pre-
mium paid at time 0, the bonus strategy must obey the equivalence relation:
V (0) =r.
A natural question to ask here is whether there exists a situation such that
the constraint X(n) =0 again secures that the prospective quantities coincide
with the retrospective quantities, so saving us from a discussion on future
bonus strategies. Before we continue, we must ask ourselves whether there
exists an investment strategy such that we can construct a bonus strategy
leading to X(n) = 0? The answer is yes, obviously, since the investment
r =0 brings us back to the situation in Chapter 2. In the following, we give a
sufficient condition on r, less trivial than r =0, for a bonus strategy leading
to X(n) =0. Let us check the consequence of the constraint. We have that
X(n) =0 ⇒ (4.5)
l (n) =V

(n) =U (n) ⇒
V (t) =v
n−t
E
Q
t
|U (n)] −ro
n−t
=v
n−t

v
−(n−t)

U (t) +ro
n−t

−ro
n−t
=U (t) .
In the third line we use the fact that the interest on U (t) and the premiums
paid to U over (t. n] under Q are expected to equal r. This can be seen since
Q is exactly constructed such that
E
Q
|Z] =r;
see Equation (4.4). In Chapter 2 we obtained, for risk-free investments, the
following:
X(n) =0 ⇒V (t) =U (t) .
The implications following Equation (4.5) generalize this result to the situation
with risky investments.
A sufficient condition on r for X(n) = 0 is that we do not risk losing
parts of V
g
via risky investments, no matter what the outcome of Z. We can
write this using an inequality:
U (t. J) −V
g
(t. J) ≥0.
108 Bonus, binomial and Black–Scholes
where the argument J means that the outcome Z(t) =J is plugged in.
If the constraint X(n) =0 is fulfilled, we have seen that V (t) =U (t). If
we add the classical solvency rule X(t) ≥ 0, a sufficient condition on r is
that we do not risk losing parts of V

via risky investments, no matter that
the outcome of Z. We can write this using an inequality:
X(t. J) ≥0.
where, again, the argument J means that the outcome Z(t) =J is plugged in.
The conditions limit the investments in risky assets. However, one may
wish to disregard these conditions. One can formalize a general investment
strategy where the proportion of funds invested in risky assets, r(t) S (t −1) ¡
X(t −1), is a function O of (t. V

(t −1) . X(t −1)), i.e.
r(t) S (t −1)
X(t −1)
=O(t. V

(t −1) . X(t −1)) .
Then,
X(t) = (1+r +O(t. V

(t −1) . X(t −1)) (Z(t) −r)) X(t −1)
+c (t) −o(t) .
One can for example think of the special case where the value of investments
in S is a constant proportion 0 of U:
r(t) S (t −1) =0U (t −1) ⇔
O(t. V

(t −1) . X(t −1)) =
r(t) S (t −1)
X(t −1)
=
0U (t −1)
X(t −1)
=0
V

(t −1) +X(t −1)
X(t −1)
.
If one disregards the condition X(n) =0 in the market with risky assets,
the future bonus strategy is inevitably brought into the market values of
unguaranteed payments. Then a natural question is: how is the bonus strategy
linked to the financial market? One possibility is to suggest an appropriate
explicit link which makes sense both from an economical and a mathematical
point of view. Many articles on the subject, introducing financial valuation
in life insurance with bonus, take as their starting points a certain explicit
specification of the link and the authors then consider quantities like that
gives in Equation (4.3).
4.2 Discrete-time insurance model 109
It is important to discuss how the bonus strategy is currently decided, but
just as important is to emphasize that there are no true or false answers to
the above question. Certainly, there exist answers which do not conform with
legislation but, on the other hand, legislation usually gives some degrees of
freedom. What we can show is that, given our simple financial market and one
qualitative specification of the link between bonus and the financial market,
there are no quantitative degrees of freedom if the bonus strategy has to be
fair from a financial point of view.
We take as our starting point the fact that the size of X determines the
bonus strategy. We let the bonus interest rate be linked to X such that it is
determined by
r
o
(t) =r

(t) +
4(t. X(t −1))
V

(t −1)
.
where 4 is a function specifying how the bonus interest depends on X.
Hereby,
c (t) −o(t) =

r (t) −r
o
(t)

V

(t −1)
=

r (t) −r

(t) −
4(t. X(t −1))
V

(t −1)

V

(t −1)
=c (t) −4(t. X(t −1)) .
One can think of a function 4 in the following form:
4(t. x) =(8(t) +o(t) x)
+
. (4.6)
where o and 8 are deterministic functions of t. In Section 4.2.1 we consider
a two-period example with precisely this choice of function 4. We speak of
4 as the contract function and note that a contract function in the form given
by Equation (4.6) contains two interesting constructions as special cases. If
8(t) =0 and o(t) =o >0, we obtain
4(t. X(t −1)) =o(X(t −1))
+
.
i.e. that the bonus is set as a constant part o of the undistributed reserve if
this is positive.
If o(t) =1 and 8(t) =−K -0, we obtain
4(t. X(t −1)) =(X(t −1) −K)
+
.
i.e. that the bonus is set such that the part of the undistributed reserve which
exceeds a constant buffer, K, is paid out.
110 Bonus, binomial and Black–Scholes
4.2.1 Two-period model
In this section we consider a contract over two time periods and discuss
bonus interest rates and investments. We start out by giving in Figure 4.1 an
overview of the different quantities introduced in the previous section in the
two periods. Hereafter, we specify the connection between the bonus interest
rate and the financial market.
We assume that the company at time 0 decides upon a bonus interest
rate for the first year that does not depend on the capital gains in that year.
Hereby, r
o
(1) becomes independent of Z(1). Correspondingly, we let the
bonus interest rate in the second year be independent of the capital gains
in that year. However, we let the bonus interest rate in the second year
depend on the capital gains in the first year. Then we can write that r
o
(2)
becomes dependent on Z(1) but independent of Z(2). The consequence is
that (see Figure 4.1) V

(1) and l (1) are independent of Z(1), whereas l (2)
is independent of Z(2).
This accords with the conception of the bonus interest rate as an interest
rate which is decided for the subsequent year, independently of the capital
gains in that year but dependent on the capital gains in the preceding years.
In that sense, the bonus interest rate is always “a year behind” the capital
gains. Because of this construction, we need two periods for an illustration:
one period to earn capital gains and one period for the redistribution. A con-
struction where the capital gains are redistributed in the same year as they
are earned could have been illustrated using a one-period model.
We let the bonus interest rate in the second year depend on the capital
gains in the first year through the following formula:
r
o
(2) =r

(2) +
(8+oX(1))
+
V

(1)
.
S(0)
V

(0)
U(0)
b(0)
V
g
(0)
V
b
(0)
V(0)

S(1)
V

(1)
U(1)
b(1)
V
g
(1)
V
b
(1)
V(1)
→ [V

(2)] = [b (2)]
Figure 4.1. A two-period insurance model.
4.2 Discrete-time insurance model 111
and we obtain the following expression for the technical reserve after the
second year:
V

(2) =

2+r
o
(1)

1+r
o
(2)

r.
which only depends on the capital gains in the first year. As a curiosity, we
derive the market value of the unguaranteed payments after the first year, the
so-called bonus potential, as follows:
V
b
(1) =
E
Q
1
|V

(2)] −l (1)
1+r
=

2+r
o
(1)

1+r
o
(2)

r−l (1)
1+r
=

2+r
o
(1)

1+r
o
(2)

r−V

(1) (1+r

(2))
1+r
=

2+r
o
(1)

1+r
o
(2)

r−

2+r
o
(1)

r(1+r

(2))
1+r
=

2+r
o
(1)

r
o
(2) −r

(2)

1+r
r
=
(8+oX(1))
+
1+r
.
We see that this simple construction of bonus interest rates corresponds to
a situation where the market value of unguaranteed payments after the first
year is simply a part of the undistributed reserve.
Hereafter, we are interested in a connection between a fair investment r
and a fair pair of parameters in the bonus strategy (o. 8). We determine this
connection simply by the equivalence relation
V (0) =r.
which yields
1
(1+r)
2
E
Q
|V

(2)] −
r
1+r
=r ⇔

2+r
o
(1)

1+E
Q
¸
r
o
(2)
¸
=1+r +(1+r)
2

2+r
o
(1)

1+q (u) r
o
(2. u) +q (J) r
o
(2. J)

=(1+r) (2+r) .
Here, r
o
(2. u) and r
o
(2. J) are the bonus interest rates in year 2 corresponding
to the stock going up and down in year 1, respectively. If
8+oX(1. J) -0 -8+oX(1. u) . (4.7)
112 Bonus, binomial and Black–Scholes
we have that r
o
(2. u) > r

= r
o
(2. J), such that a bonus is paid out in the
second year if and only if the risky asset goes up in the first year, and
furthermore we have
( 2 + r
o
(1) )

1+r

+q (u)

8+oX(1. u)
V

(1)

= (1+r) (2+r) ⇔
( 2 + r
o
(1) )

1+r

+q (u)
8+o

r −r
o
(1) +(rs¡r) (u−r)

(2+r
o
(1))

= (1+r) (2+r) .
which implies that
q ( u )

8+o

r −r
o
(1) +
rs
r
(u−r)

=(1+r) (2+r) −

2+r
o
(1)

(1+r

) . (4.8)
Equation(4.8) candeterminefair combinationsof thethreeparameters (o. 8. r),
and we then have to verify Equation (4.7). At least, this is what we claim at this
moment because we still have not presented the argument for the application
of the artificial probability measure Q, but we shall come back to this later.
In the rest of this section, we consider a numerical example where we let
s =1 and r =1. We let 8 =0 such that the bonus interest rate is determined
exclusively by the parameter o. The interest rate r is 5%, the bonus interest
rate in year 1 is also 5%, while the guaranteed interest rate is 3%. The
risky asset can raise by 20% or fall by 10%, i.e. u =20%, J =−10%. With
these numbers we find that q (u) =q (J) =1¡2, i.e. we determine the market
values under a probability measure where the risky asset both rises and falls
with probability 1/2. Before determination of o and r, we can readily verify
Equation (4.7). Since r
o
(1) = r, we have that c (1) −o(1) = 0 such that
X(1) =rs (Z(t) −r). But with 8 =0, Equation (4.7) is then obviously true.
Equation (4.8) now gives a fair connection between o and r:
or =0.55.
Thus, we have that the redistribution parameter o is a hyperbola as a function
of the investment parameter r. We see that if our proportion of the risky asset
is less than 55%, we have o >1. A redistribution parameter larger than unity
makes sense since there is a bonus potential in the second year which must
be redistributed but which is not included in X(1).
In Figure 4.2 we find all quantities for the case where we have placed
60% in the risky asset, i.e. r = 0.6. This gives a redistribution parameter o
4.2 Discrete-time insurance model 113
S(1) = 1.2
V

(1) = 2.05
b(1) = 2.11
V
g
(1) = 2.01
V
b
(1) = 0.08
V(1) = 2.09
[b(2) = 2.19]
S(0) = 1
V

(0) = 1
X(0) = 0
b(0) = 2.09
V
g
(0) = 0.94
V
b
(0) = 0.06
V(0) = 1
S(1) = 0.9
V

(1) = 2.05
X(1) = −0.09
b(1) = 2.11
V
g
(1) = 2.01
V
b
(1) = 0
V(1) = 2.01
[b(2) = 2.11]
X(1) = 0.09
Figure 4.2. Two-period insurance model.
of approximately 0.91. Figure 4.2 shows the outcomes of all quantities in
Figure 4.1, depending on whether the risky asset goes up or down in the first
period.
We see that the technical reserve after the first year is 2.05, independent of
the outcome of the risky asset in that year. In contrast, the development of the
technical reserve after the second year, and thus the terminal payment, equals
2.19 or 2.11, depending on the outcome of the risky asset. If the risky asset
goes up in the first year, the bonus interest rate in the second year becomes
7%. If the risky asset goes down in the first year, the bonus interest rate in
the second year equals the first order interest rate, 3%. If the risky asset goes
up in the first year, both the individual and the collective bonus potential are
0.04 after the first year. If the risky asset goes down in the first year, all safety
margins are lost and both bonus potentials are zero. The difference between
the bonus potential after the first year in the two scenarios pays the difference
between the two benefits paid out after two years.
So far we have simply claimed how to calculate all values. We now give
a brief argument for these values as they look in the example above. The
argument simply shows how it is possible, by tactical investments of the
premiums, precisely to obtain the benefits that are due after the second year.
Hereby, the values above become natural as reserves for future obligations.
114 Bonus, binomial and Black–Scholes
We calculate the investment which is required after the first year if the pay-
ment after the second year has to be met. There are two cases corresponding
to the risky asset going up or down in the first year.
Consider the case where the risky asset goes up. Now, the risky asset can
either go up or down in the second year. No matter what happens to the risky
asset in the second year, we need 2.19 after the second year. If we let h
1
(2. u)
denote the number of risky assets and h
0
(2. u) denote the investment in the
risk-free interest rate, we obtain the following equation system:
h
1
(2. u) 1.44+h
0
(2. u) 1.10 =2.19.
h
1
(2. u) 1.08+h
0
(2. u) 1.10 =2.19.
where 1.44 and 1.08 are the prices of the risky asset after the second year
depending on the outcome of the risky asset in the second year and given that
the asset goes up in the first year. The number 1.10 is (1.05)
2
rounded off,
which is the value after two years of one unit at time 0 including accumulated
interest rate by 5%. The solution is given by
h
1
(2. u) =0. h
0
(2. u) =1.99.
and we note that the price of this investment is precisely given by
1.99×1.05 =2.09 =V (1. u) .
A corresponding system is obtained if the asset goes down in the first year:
h
1
(2. J) 1.08+h
0
(2. J) 1.10 =2.11.
h
1
(2. J) 0.81+h
0
(2. J) 1.10 =2.11.
which has the following solution:
h
1
(2. J) =0. h
0
(2. J) =1.92.
and the price for this investment is given by
1.92×1.05 =2.01 =V (1. J) .
As one would expect, we shall not invest in the risky asset in the second year,
no matter its outcome in the first, since the bonus interest rate in the second
year is independent of the outcome of the risky asset in that year.
Now we know what to do after the first year no matter the outcome in the
first year, but what should we do at time 0? Here again, we can set up a system
where we are trying to hit what it takes to establish the required portfolios after
the first year, no matter the outcome in the first year. We know, however, that
4.3 The binomial model 115
in addition to the capital gains after the first year we also receive a premium
payment of one, such that the equation system becomes
h
1
(1) 1.20+h
0
(1) 1.05+1 =2.09.
h
1
(1) 0.90+h
0
(1) 1.05+1 =2.01.
This system has the following solution:
h
0
(1) =0.74. h
1
(1) =0.26.
and, finally, we see that this investment has exactly the value r:
0.74×1+0.26×1 =1 =r.
The price of this investment is exactly what is paid as a premium at time 0,
and we have thus constructed a portfolio strategy with which we can meet
our obligations without taking any risk. The arbitrage argument states that
the market value of the obligation must be the price of establishing such a
portfolio with which we can precisely meet the obligations. Thus, we conclude
the numbers in Figure 4.2.
The question is now: was this magic? The boring answer is as follows.
For the calculation of all quantities, we used the probabilities q (u) and
q (J). These seemed to be drawn from a hat, but we must admit that they
were constructed precisely such that the above calculations work out neatly.
Hereby, we mean that by calculating expected values with the probability
measure Q, we obtain values with which we can meet the obligations without
risk. In the next section, we take a closer look at the construction of the
measure Q and other concepts in financial mathematics.
4.3 The binomial model
4.3.1 The one-period model
In this section we consider the one-period version of the simplest possible non-
trivial financial market, where the stock after one period takes one value out of
only two: not because we think that this market is very realistic, but because
it is very simple to understand the concepts of financial mathematics in this
market. The calculations boil down to two equations with two unknowns and,
furthermore, the model is often applied in practice.
116 Bonus, binomial and Black–Scholes
The model consists of two investment possibilities: a bond and a stock.
The price of the bond is deterministic and takes the following values:
S
0
(0) =1.
S
0
(1) =(1+r) S
0
(0) =1+r.
where r is a deterministic interest rate for the period 0 to 1. We can also
interpret the investment opportunity S
0
as the possibility of saving money
on a bank account with interest rate r. The price of the stock is a stochastic
process and takes the following values:
S
1
(0) =s.
S
1
(1) =s (1+Z) .
Z =
¸
u with probability ¡
u
.
J with probability ¡
J
.
We study the dynamics of special investment portfolios in the market and
define a portfolio (strategy) as a vector h =

h
0
. h
1

which describes the
number of bonds and stocks, respectively, in the portfolio at time 0. We
allow for negative components in the portfolio and interpret this as having
sold the assets. Such a position is called short-selling. The interpretation of
short-selling of bonds is that we borrow money in the bank at interest rate r,
whereas the interpretation of short-selling of stocks is that we, against a
payment at time 0, oblige ourselves to pay the price of the stock in the future.
Linked to a portfolio h, we define a value process as follows:
V (t. h) =h
0
S
0
(t) +h
1
S
1
(t) . t =0. 1.
The notion of arbitrage plays a very important role, and we define an
arbitrage portfolio as a portfolio obeying the following relations:
V (0. h) =0.
P (V (1. h) ≥0) =1.
P (V (1. h) >0) >0.
(4.9)
Thus, an arbitrage portfolio is a portfolio which is established at zero price
at time 0, has a non-negative value at time 1, but with a chance of being
positive. It is important here to emphasize that we usually require from a
market that there exist no arbitrage portfolios, and we then speak of the
market as arbitrage-free. Thus, for a given market one can pose the question,
4.3 The binomial model 117
is this market arbitrage-free? It is easy to see that the binomial market is
arbitrage-free if
J -r -u.
The solution (q (u) . q (J)) to the following system of equations:
q (u) u+q (J) J =r. (4.10)
q (u) +q (J) =1. (4.11)
becomes
q (u) =
r −J
u−J
.
q (J) =
u−r
u−J
.
(4.12)
If and only if we have the inequalities J -r -u, then q (u), q (J) >0 and the
relation q (u) +q (J) =1 in Equation (4.11) assures us that we can interpret
q (u) and q (J) as probabilities under a measure that we denote by Q. For
this measure we have, according to Equation (4.10), the following:
1
1+r
E
Q
¸
S
1
(1)
¸
=
1
1+r
(q (u) s (1+u) +q (J) s (1+J))
=s
1+q (u) u+q (u) J
1+r
=s.
We see that the expectation of the discounted stock value at time 1 equals the
stock price at time 0. Aprobability measure under which this is the case is called
a martingale measure (or a risk-neutral measure or a risk-adjusted measure).
From the reasoning above, we conclude that a binomial model

S
0
. S
1

is arbitrage-free if and only if there exists a martingale measure. This is
an important result and, in fact, such a link between no arbitrage and the
existence of a martingale measure goes far beyond the binomial model.
We assume that the market is arbitrage-free, i.e. J - r - u, and we study
the valuation problem for so-called contingent claims. It is important to note
that whereas arbitrage is a matter of what one cannot obtain in the market,
valuation is a matter of what one can obtain.
A contingent claim is a stochastic variable in the form X = 4(Z). Thus,
it is a stochastic variable since Z is a stochastic variable. Equivalently, it is
determined by the outcome of S
1
(1) exclusively. The function 4 is called
the contract function.
A typical example of a contingent claim is a European call option on the
stock with strike price K. A European call option gives the right, but not the
118 Bonus, binomial and Black–Scholes
obligation, to buy a stock at time 1 at price K. In our model this contract
is only interesting if s (1+J) - K - s (1+u). If S
1
(1) > K, the option is
exercised, and the owner pays K for the stock and sells it on the market for
s (1+u) with a gain equal to s (1+u) −K. If S
1
(1) -K, the owner gives up
the option with a gain of 0; i.e.
X =4(Z) =(s (1+Z) −K)
+
=
¸
s (1+u) −K. Z =u.
0. Z =J.
(4.13)
Our problem is to find a fair price of this contract if one exists. The price at
time t is denoted by H(t. X), and we easily see that in order to avoid arbitrage
we must have that H(1. X) =X; the problem then is to find H(0. X).
A contingent claim is called attainable if there exists a portfolio h such
that
V (1. h) =X. (4.14)
In that case, h is called a hedging or replicating portfolio. If all contingent
claims are attainable, the market is said to be complete. If a claim is attainable
with the hedging portfolio h, there is in reality no difference between holding
the claim and holding the hedging portfolio. Then it is easy to see that in
order to avoid arbitrage possibilities in a market including the claim, the price
of the claim at time 0 must equal the price of establishing a hedging portfolio
at time 0, i.e.
H(0. X) =V (0. h) .
If the market is complete, all claims can be priced by determining the price
of the hedging portfolio. A natural question is now whether the one-period
binomial market is complete.
According to Equation (4.14), we know that for a portfolio hedging X,
h
0
(1+r) +h
1
s (1+u) =4(u) .
h
0
(1+r) +h
1
s (1+J) =4(J) .
(4.15)
This is a system of two equations with two unknowns

h
0
. h
1

, the solution
of which, under the assumption J -r -u, is determined as follows:
h
1
=
4(u) −4(J)
s (u−J)
. (4.16)
h
0
=
1
1+r
(1+u) 4(J) −(1+J) 4(u)
u−J
.
4.3 The binomial model 119
We can now calculate the price H(0. X) =V (0. h):
V (0. h) =h
0
+h
1
s
=
1
1+r
(1+u) 4(J) −(1+J) 4(u)
u−J
+
4(u) −4(J)
(u−J)
=
1
1+r
(1+u) 4(J) −(1+J) 4(u) +(1+r) 4(u) −(1+r) 4(J)
u−J
=
1
1+r
(q (u) 4(u) +q (J) 4(J)) . (4.17)
where q (u) and q (J) are given by Equation (4.12). We conclude that the
one-period binomial model is complete and that X =4(Z) at time 0 has the
price given by Equation (4.17) with (q (u) . q (J)) given by Equation (4.12).
We see that the price of a contingent claim X can be calculated by the
so-called risk-neutral formula, as follows:
H(0. X) =
1
1+r
E
Q
|X] .
where Q, given by the probabilities (q (u) . q (J)), is the unique probability
measure under which also the price of the risky asset can be calculated
using a risk-neutral formula. By risk neutrality we mean the intuitive quality
that prices are calculated simply as expected discounted claims. However,
it is important to realize that the pricing formula does not assume risk-
neutral agents. Risk neutrality is just a notion which shows up naturally when
interpreting the quantities q (u) and q (J) as probabilities.
It may be surprising that the probabilities ¡(u) and ¡(J) do not play any
important role in the pricing formula. After all, if we hold the option, the
expected gain is affected by the objective probability measure. On the other
hand, by buying the option we give up the possibility of investing the price of
the option on the same market with corresponding objective changes in prices.
The calculations above show that these two circumstances “level out” and
that the disappearance of ¡(u) and ¡(J) does not seem that counter-intuitive
after all.
We end this section with a numerical example, in which we find the price
of a European option, Equation (4.13), with the parameters r =5%, u =20%,
J = −10%, K = 110, ¡(u) = 0.6, ¡(J) = 0.4. The first idea could be to
calculate the expected present value under the objective measure to obtain
1
1.05
E
P
|4(Z)] =
1
1.05
(10¡(u) +0¡(J)) =5.71.
120 Bonus, binomial and Black–Scholes
S
0
(1) = 1.05
S
1
(1) = 120
∏(1) = 10
S
0
(0) = 1
S
1
(0) = 100
∏(0) = 4.76
h
0
= −28.57
h
1
= 0.33
S
0
(1) = 1.05
S
1
(1) = 90
∏(1) = 0
Figure 4.3. One-period option model.
One may suggest loading that price according to, for example, the variance
or the standard deviation. We illustrate the completely different consequences
of an arbitrage argument in Figure 4.3.
Figure 4.3 shows how the dynamics of the risk-free asset, the risky asset
and the option depend on the outcome of the stochastic variable Z. Firstly,
we write S
0
and S
1
for the two time points 0 and 1. Hereafter, we write the
price of the option at time 1, which is given by Equation (4.13). Finally, we
calculate H(0) and h from Equations (4.17) and (4.16), which in this example
are as follows:
H(0) =
1
1.05
E
Q
|4(Z)]
=
1
1.05
(q (u) 4(u) +q (J) 4(J))
=
1
1.05

0.05−(−0.1)
0.2−(−0.1)
10+
0.2−0.05
0.2−(−0.1)
0

=4.76;
h
1
=
10−0
100(0.2−(−0.1))
=0.33;
h
0
=
1
1.05
1.2×0−0.9×10
0.2−(−0.1)
=−28.57.
We can now verify that the price for establishing the portfolio h is exactly
given by
V (0. h) =−28.57+0.33×100 =4.76.
whereas the value of the portfolio at time 1 becomes
V (1. h) =
¸
−28.57×1.05+0.33×120 =10. Z =u.
−28.57×1.05+0.33×90 =0. Z =J.
4.3 The binomial model 121
4.3.2 The multi-period model
A drawback of the one-period model is, of course, the very simple asset
model with only two possible outcomes. If we simply increase the number
of outcomes and then search for replicating portfolios, we get into trouble
since the number of unknowns in Equation (4.15) will still be two,

h
0
. h
1

,
whereas the number of equations will equal the number of possible outcomes.
Thus, the system will only have a solution for special trivial claims such as
4(Z) = S
1
(1). We can say that the completeness of the binomial model
exactly connects to the fact that the number of equations in the system shown
in Equation (4.15) equals the number of unknowns.
The answer is to allow for intermediary trading in the two assets. If we put
a number of one-period models in a row, the number of outcomes increases
with the number of periods. If, on the other hand, we allow agents to change
their portfolio after each one-period model, the completeness is maintained.
We assume a time horizon n and define the price processes for the risk-free
asset and the risky asset as follows:
S
0
(t) =(1+r) S
0
(t −1) . t =1. . . . . n.
S
0
(0) =1.
S
1
(t) =(1+Z(t)) S
1
(t −1) . t =1. . . . . n.
S
1
(0) =s.
where Z(1) . . . . . Z(n) are independent stochastic variables which take one
out of two values u and J with the probabilities ¡(u) and ¡(J), respectively.
A portfolio (strategy) is a process
h =]h(t)]
t=1.... .n
=
¸
h
0
(t) . h
1
(t))]
t=1.... .n
.
such that h(t) is a function of

S
1
(1) . . . . . S
1
(t −1)

. The value process
linked to the portfolio h is defined by
V (t. h) =h
0
(t) S
0
(t) +h
1
(t) S
1
(t) . t =0. . . . . n.
The portfolio at time t, h(t), is the portfolio held during the period |t −1. t].
It is intuitively clear that this portfolio only can depend on the prices until
time t −1 and not on the future prices from time t and onwards. Otherwise,
we would of course take advantage of the future price changes when deciding
the portfolio today. We say that the portfolio must be predictable. Particularly
122 Bonus, binomial and Black–Scholes
interesting are the self-financing portfolios, for which
h
0
(t) S
0
(t) +h
1
(t) S
1
(t) =h
0
(t +1) S
0
(t) +h
1
(t +1) S
1
(t) .
t =1. . . . . n−1. (4.18)
For a self-financing portfolio, no capital is injected or withdrawn at any of the
time points t =1. . . . . n−1, and all new investments at time t are financed by
capital gains on old investments. Hereafter, we define an arbitrage portfolio
as a self-financing portfolio for which Equation (4.9) holds with the time
point 1 replaced by n.
The martingale measure is now given by the positive probabilities (q (u) .
q (J)) for which the discounted price process for the risky asset is a martin-
gale, i.e.
1
1+r
E
Q
¸
S
1
(t +1)

S
1
(t) =s
¸
=s.
and we easily see that (q (u) . q (J)) again is given by Equation (4.12). In
order for such a set of probabilities to exist, we must require, as in the one-
period model, that J - r - u. As in the one-period model, this condition
makes sure that it is not possible to construct arbitrage portfolios. Thus, the
multi-period model is also arbitrage-free exactly if there exists a martingale
measure.
A simple contingent claim is a stochastic variable X =4

S
1
(n)

, whereas
more general claims are given in the form X = 4

S
1
(1) . . . . . S
1
(n)

. We
note here that a claim which is due at time t and which depends on S
1
(t)
or S
1
(1) . . . . . S
1
(t), respectively, constitutes a simple or a general claim,
respectively, in the partial model running from time 0 until time t.
A claim is called attainable if there exists a self-financing portfolio h such
that V (n. h) = X, and h is then called hedging or replicating. There is, in
principle, no difference between holding an attainable claim and the portfolio
that hedges the claim, and, thus, the only reasonable price of such a claim is
exactly the value of the hedging portfolio, i.e.
H(t. X) =V (t. h) .
By “reasonable price” we mean that if the price is any different, it is possible to
construct arbitrage portfolios. We speak, in that case, simply of the arbitrage-
free price of the claim. If all claims in a market are attainable, the market
is said to be complete, and we can ask whether the multi-period model is
complete. By copying the argument in the one-period model for each period,
one can show that the multi-period model is also complete and that prices
and hedging portfolios can be found by solving a number of systems of two
4.3 The binomial model 123
equations with two unknowns. We shall not carry out the argument in detail,
but instead illustrate the technique for a specific claim, a European call option
in a three-period model.
During the rest of this section, we want to find the price of a European
option with the parameters r =5%, u =20%, J =−10%, K =110. Figure 4.4
shows how the dynamics of the prices of the risk-free asset, the risky asset
and the option depend on the outcome of the stochastic variables Z(1), Z(2)
and Z(3). We have that q (u) = q (J) = 1¡2. Firstly, we calculate S
0
and
S
1
for the time points 0, 1, 2 and 3. Hereafter, we write the option price at
time 3, which is now given by the terminal condition
H(3. X) =V (3. h) =X =

S
1
(3) −K

+
.
S
0
(3) = 1.16
S
1
(3) = 172.8
∏(3) = 39.24
S
0
(2) = 1.10
S
1
(2) = 144
∏(2) = 39.24
h
0
(3) = −95.02
h
1
(3) = 1
S
0
(1) = 1.05
S
1
(1) = 120
∏(1) = 23.13
h
0
(2) = −72.91
h
1
(2) = 0.83
S
0
(3) = 1.16
S
1
(3) = 129.6
∏(3) = 19.6
S
0
(0) = 1
S
1
(0) = 100
∏(0) = 13.13
h
0
(1) = −49.15
h
1
(1) = 0.62
S
0
(2) = 1.10
S
1
(2) = 108
∏(2) = 9.33
h
0
(3) = −50.79
h
1
(3) = 0.60
S
0
(1) = 1.05
S
1
(1) = 90
∏(1) = 4.44
h
0
(2) = −25.40
h
1
(2) = 0.35
S
0
(3) = 1.16
S
1
(3) = 97.2
∏(3) = 0
S
0
(2) = 1.10
S
1
(2) = 81
∏(2) = 0
h
0
(3) = 0
h
1
(3) = 0
S
0
(3) = 1.16
S
1
(3) = 72.9
∏(3) = 0
Figure 4.4. Multi-period option model.
124 Bonus, binomial and Black–Scholes
Now we calculate H(2. X) and h(3) using Equations (4.17) and (4.16)
for each of the three possible outcomes at time 2. For the first outcome,
corresponding to the risky asset going up in the two first periods and at time
2 taking the value 144, we obtain
H(2. X) =
1
1.05
E
Q
|4(Z)] =
1
1.05

62.8×
1
2
+19.6×
1
2

=39.24.
h
1
(3) =
62.8−19.6
144(0.2−(−0.1))
=1.
h
0
(3) =
1
(1.05)
3
1.2×19.6−0.9×62.8
0.2−(−0.1)
=−95.02.
We can now verify that the price of establishing the portfolio h is exactly
given by
V (2. h) =−95×(1.05)
2
+1×144 =39.24.
whereas the value of the portfolio at time 3 becomes
V (3. h) =
¸
−95.02×(1.05)
3
+1×172.8 =62.8. Z(3) =u.
−95.02×(1.05)
3
+1×129.6 =19.6. Z(3) =J.
Now, V (2. h) can substitute for H(2). We continue in the same way by
filling in H(2) and h(3) for the other possible outcomes of the risky asset
at time 2, 108 and 81. Hereafter, we go to time point 1 and fill in H(1) and
h(2), where now H(2) makes up the claim in a new one-year calculation.
Finally, we find H(0). It is very simple to program the scheme described
here since the calculations at each point in the tree are the same.
4.4 The Black–Scholes model
In this section we study the most important and well known of all continuous-
time models, namely the model suggested by Black, Scholes and Merton in
1973 and awarded the Nobel Prize in 1997. The model is not particularly
realistic, but it provides a good approximation to reality on short time horizons.
Furthermore, the model provides access to simple calculations of unique
prices, and this makes the model important. Another argument is that a very
large part of financial theory takes the Black–Scholes model as its starting
point, and obviously, in order to understand results based on generalizations
of the Black–Scholes model, we first have to understand the results based
on the Black–Scholes model itself. For further studies of the Black–Scholes
model, we refer to Björk (1994, 2004).
4.4 The Black–Scholes model 125
The Black–Scholes model consists of two assets with dynamics given by
dS
0
(t) =rS
0
(t) dt. (4.19)
S
0
(0) =1.
dS
1
(t) =oS
1
(t) dt +uS
1
(t) dW (t) . (4.20)
S
1
(0) =s.
where r, o and u are deterministic constants. It is possible to generalize the
model to the situation where r, o and u are functions of

t. S
1
(t)

, but such
a generalization provides no further insight. We briefly mention some other
possible generalizations in Section 4.6.
The idea of the dynamics of S
0
is clear since Equation (4.19) simply states
that S
0
fulfils a deterministic differential equation with the following solution:
d
dt
S
0
(t) =rS
0
(t) . S
0
0
=1.
S
0
(t) =e
rt
.
We can interpret investing in S
0
as depositing on a bank account at a con-
stant interest rate. The dynamics of S
1
requires comment. It is based on the
assumption that a change of asset price in a “small” time interval (t. t +At]
can be approximated as follows:
AS (t) = S (t +At) −S (t) (4.21)
≈ oS (t) At +uS (t) (W (t +At) −W (t))
= oS (t) At +uS (t) AW (t) .
where W is a so-called standard Brownian motion or Wiener process. Such
a Wiener process is characterized by continuity and independent normally
distributed increments such that
AW (t) ∼N (0. At) .
Actually, Equation (4.21) is only assumed for infinitely small time intervals,
and this is written in the form given by Equation (4.20). In spite of the fact that
W is an extremely complicated process to work with (for example, though
continuous it is not differentiable in any point which makes it impossible
to illustrate), it plays an important role in various disciplines of applied
mathematics, including mathematical finance.
We now continue by defining various concepts in mathematical finance.
The content of these concepts correspond, broadly speaking, to the content
126 Bonus, binomial and Black–Scholes
they were given in the discrete-time models. A portfolio (strategy) is a pro-
cess h = ]h(t)]
t∈|0.n]
=
¸
h
0
(t) . h
1
(t)
¸
t∈|0.n]
such that h(t) is a function
of

S
1
(s) . 0 ≤s ≤t

. The value process linked to the portfolio h is now
defined by
V (t. h) =h
0
(t) S
0
(t) +h
1
(t) S
1
(t) . t ∈ |0. n] .
The portfolio at a given point in time t, h(t), is interpreted as the portfolio
which is held during the period |t. t +dt). It is intuitively clear that this
portfolio is allowed to depend on the prices until time t and not on the future
prices after time t. Otherwise, one could arrange a portfolio based on future
price changes. Of particular interest are the self-financing portfolios for which
dV (t. h) = h
0
(t) dS
0
(t) +h
1
(t) dS
1
(t) (4.22)
= rV (t. h) dt +h
1
(t) S
1
(t) ((o−r) dt +u dW (t)) . (4.23)
We do not inject money in or withdraw money from these portfolios at any
point in time, and all new investments at time t are financed by capital gains
on old investments. It is by no means evident that the idea behind being
self-financing is formalized by the definition given in Equation (4.22), but a
discrete- time argument and an appropriate content given to Equation (4.20)
leads to that result. Hereafter, an arbitrage portfolio is defined as a portfolio
with the following qualities:
V (0. h) =0.
P (V (n. h) ≥0) =1.
P (V (n. h) >0) >0.
As in the discrete-time models, we are interested in prices on deriva-
tives, and again we define a simple contingent claim as a stochastic vari-
able X = 4

S
1
(n)

, whereas more general claims are in the form X =
4

S
1
(t) . t ∈ |0. n]

. As in the binomial model, it is now natural to discuss
concepts such as martingale measures, attainable claims and complete mar-
kets. However, a rigorous discussion requires knowledge of the results for
stochastic processes which lie beyond the purpose of this material and we
refer to Björk (1994). We can, however, take a few steps further, by follow-
ing the reasoning of Black, Scholes and Merton, who did well with a limited
amount of advanced results for Wiener processes.
A claim is called attainable if there exists a self- financing portfolio h
such that
V (n. h) =X. (4.24)
4.4 The Black–Scholes model 127
and h is called the hedging or replicating portfolio. There is no difference
between holding an attainable claim and the hedging portfolio. Therefore,
the only arbitrage-free price for such a claim is the value of the hedging
portfolio, i.e.
H(t. X) =V (t. h) . (4.25)
If all claims in a market are attainable, the market is said to be complete, and
it is now natural to ask whether the Black–Scholes market model is complete.
If the answer is yes, we have found the arbitrage-free price on every claim at
any point in time through Equation (4.25), where h is the hedging portfolio.
To show that the Black–Scholes market is complete can, in advance, seem
impossible. For the binomial model the task was reduced to counting equations
and unknowns and realizing that the system gave us both hedging portfolios
and prices. With the introduction of the Wiener process as the stochastic
element, this looks difficult. However, in spite of the irregular behavior of
the Wiener process, there are a number of quite simple rules of calculation
which we use in the following. We shall not go into detail, but just use
them as we need them. However, we should emphasize that these rules of
calculation are not based on any financial reasoning but are pure mathematical
and probability theoretical results about the Wiener process.
Now, we let H(t) denote the price at time t for a simple claim X, for which
we seek a hedging portfolio h and an arbitrage-free price H. Now we assume
(a non-trivial assumption) that the price of the claim X can be written as a
function of time and the price of the stock, i.e. H(t) = E

t. S
1
(t)

, where
E (t. s) is a deterministic function of two variables. We mention that the
function E can of course have derivatives with respect to the two variables,
and we assume that such derivatives exist to the extent that we need them
(another non-trivial assumption).
Thus, we seek a pair (h. H) such that Equations (4.24) and (4.25) are
fulfilled. Here, we need the dynamics of various price processes. The dynam-
ics of the prices S
0
and S
1
are known, but what about the dynamics of H?
Appropriately, there exists a deep result about how the dynamics of S
1
, Equa-
tion (4.20), is reflected in a sufficiently regular function E

t. S
1
(t)

. The
result is called Itô’s formula, and its consequence is that H behaves just like
S
1
with specific processes o
H
(t) and u
H
(t) replacing o and u. The dynamics
of the price process H is given by
dH(t) =o
H
(t) H(t) dt +u
H
(t) H(t) dW (t) . (4.26)
128 Bonus, binomial and Black–Scholes
with
o
H
(t) =
oE
ot

t. S
1
(t)

+oS
1
(t)
oE
os

t. S
1
(t)

E (t. S
1
(t))
+
1
2
u
2

S
1
(t)

2
o
2
E
os
2

t. S
1
(t)

E (t. S
1
(t))
. (4.27)
u
H
(t) =
uS
1
(t)
oE
os

t. S
1
(t)

E (t. S
1
(t))
. (4.28)
On the other hand, we need a portfolio such that the value process equals
the price H, i.e., according to Equation (4.25),
H(t) =V (t. h) =h
0
(t) S
0
(t) +h
1
(t) S
1
(t) . (4.29)
and which is self-financing, i.e., according to Equation (4.22),
dH(t) = h
0
(t) dS
0
(t) +h
1
(t) dS
1
(t)
= h
0
(t) rS
0
(t) dt +h
1
(t)

oS
1
(t) dt +uS
1
(t) dW (t)

=
h
0
(t) rS
0
(t) +h
1
(t) oS
1
(t)
H(t)
H(t) dt
+
h
1
(t) uS
1
(t)
H(t)
H(t) dW (t) . (4.30)
Now we can compare Equations (4.26) and (4.30), and by equating the terms
in front of dt and dW (t) we obtain
o
H
(t) =
h
0
(t) rS
0
(t) +h
1
(t) oS
1
(t)
H(t)
.
u
H
(t) =
h
1
(t) uS
1
(t)
H(t)
.
By insertion of Equation (4.29), we obtain
o−r
u
=
o
H
(t) −r
u
H
(t)
. (4.31)
h
1
(t) =
u
H
(t) H(t)
uS
1
(t)
=
oE
os

t. S
1
(t)

. (4.32)
h
0
(t) =
H(t) −h
1
(t) S
1
(t)
S
0
(t)
. (4.33)
4.4 The Black–Scholes model 129
Equation (4.31) states that the additional expected gain beyond the risk-free
interest rate r per risk measured by the factor in front of the dW (t) term
coincides for the stock price and the price H. This quantity is called the risk
premium. If we can find H using Equation (4.31), then Equations (4.32) and
(4.33) specify the hedging portfolio.
Substituting for Equations (4.27) and (4.28) into Equation (4.31) yields
0 =
oE
ot

t. S
1
(t)

+rS
1
(t)
oE
os

t. S
1
(t)

+
1
2
u
2

S
1
(t)

2
o
2
E
os
2

t. S
1
(t)

−E

t. S
1
(t)

r. (4.34)
In addition, we have the terminal condition H(n) = 4(S (n)). This differ-
ential equation must be fulfilled at any point in time, no matter the value of
S
1
(t), and we can instead write down the corresponding deterministic partial
differential equation with terminal condition for the function E (t. s). This
differential equation triumphs as the Black–Scholes equation :
oE
ot
(t. s) +rs
oE
os
(t. s) +
1
2
u
2
s
2
o
2
E
os
2
(t. s) −E (t. s) r =0. (4.35)
E (n. s) =4(s) .
The Black–Scholes equation is a deterministic partial differential equation
which we can solve for every point (t. s). If we need the price of the claim,
we simply replace s by S
1
(t) in E (t. s). It is important to understand that the
Black–Scholes equationis adeterministictool for determinationof thestochastic
value E

t. S
1
(t)

. If, instead, we wish to carry forward this price in a small time
interval, we need the dynamics given in Equation (4.26) again. By insertion of
Equations (4.27) and (4.28) and oE

t. S
1
(t)

¡ot isolated in Equations (4.34),
we obtain the dynamics which carry forward the price H, as follows:
dH(t) =

rH(t) +(o−r) S
1
(t)
oE
os

t. S
1
(t)

dt
+uS
1
(t)
oE
os

t. S
1
(t)

dW (t) . (4.36)
By comparing Equations (4.23) and (4.36) and using Equation (4.25), we
again obtain the hedging portfolio, since h
1
(t) =oE

t. S
1
(t)

¡os, and h
0
(t)
is then determined residually from the following:
H(t) =h
0
(t) S
0
(t) +h
1
(t) S
1
(t) .
The solution to a class of deterministic partial differential equations,
including Equation (4.35), can be written as a conditional expectation in
130 Bonus, binomial and Black–Scholes
a so-called stochastic representation formula. We speak of Equation (4.20)
as the P-dynamics of S
1
, since the dynamics is presented using a process W,
which is a Wiener process under the so-called objective probability measure P.
This is the measure that relates to the investors’ expectations of the future
asset prices. Let us instead write the dynamics for S
1
using a process W
Q
,
which is a Wiener process under a different measure Q, determined such that
dS
1
(t) =rS
1
(t) dt +uS
1
(t) dW
Q
(t) . (4.37)
Under Q, we have that
E
Q
¸
e
−r(u−t)
S
1
(u)

S
1
(t) =s
¸
=s. t ≤u ≤n.
where E
Q
denotes the expectation under measure Q. We see that, except
for the discount factor e
−r(u−t)
, S
1
is a martingale under Q; Q is known as
the martingale measure. The solution to the Black–Scholes equation has the
following stochastic representation formula:
E (t. s) =E
Q
¸
e
−r(n−t)
4

S
1
(n)

S
1
(t) =s
¸
. (4.38)
For most contract functions 4, Equation (4.38) can be calculated by numer-
ical methods or by simulation methods only. However, in a few cases it can be
calculated more or less analytically; one example is the European call option,
4

S
1
(n)

=

S
1
(n) −K

+
. The distribution of S
1
under Q combined with
this simple contract function gives, after a number of tedious calculations,
a relatively explicit formula for the arbitrage-free price of a European call
option in a Black–Scholes market. This formula triumphs as the Black–Scholes
formula:
E (t. s) =sN |J
1
(t. s)] −e
−r(n−t)
KN |J
2
(t. s)] . (4.39)
Here, N is the distribution function for a standard normal distribution, and
J
1
(t. s) =
1
u

n−t

log

s
K

+

r +
1
2
u
2

(n−t)

.
J
2
(t. s) =J
1
(t. s) −u

n−t.
In the binomial model, one could wonder what happened to the objective
probabilities in the valuation formula. Correspondingly, in the Black–Scholes
model one can wonder why the expected return on the asset o does not appear
in either the Black–Scholes equation, Equation (4.35), or the Black–Scholes
formula, Equation (4.39). However, we can suggest a similar explanation.
If we hold an option in a market with a large o, we have a good chance
of making capital gains. On the other hand, we give up the possibility of
investing the price of the option in the same market with a good chance of
4.5 Continuous-time insurance model 131
capital gains. The results above show that, at the end of the day, we are
indifferent to the expected return on the stock when pricing the option.
Disregarding the presence of a martingale measure Q, one should recognize
the connection between a deterministic differential equation, a stochastic rep-
resentation formula and an explicit solution. All this is known from Thiele’s
differential equation, which is also a deterministic differential equation that
corresponds to a conditional expectation of future discounted payments, given
that the insured is alive. For most traditional versions of Thiele’s differential
equation, it is furthermore possible to write an explicit solution which is just
the reserve expressed with elementary capital values. The concrete calcula-
tions required for realizing more rigorously the connection between Thiele’s
differential equation and the Black–Scholes formula is based on the theory
for stochastic processes. Parts of Steffensen (2001) deal with this subject.
A number of differences between Thiele’s differential equation and the
Black–Scholes equation are just as important as the similarities. In Thiele’s
differential equation, we wish to find a conditional expectation under the
objective probability measure. We speak of such a conditional expectation
as a value by reference to diversification of the risk. Hereafter, we derive
a corresponding differential equation and an explicit formula. In the Black–
Scholes model, we first derive the differential equation, and we base this
derivation on the objective of replicating an attainable claim. Hereafter, it
is possible to derive a stochastic representation formula which contains an
artificial probability measure on which the explicit solution is also based.
Usually, we use the same Thiele’s differential equation for initial backward
calculation and for forward accumulation of the reserve. This works because
the stochastic element, the indicator for being alive, is constant before and
after the insurance event. Hereby, the deterministic backward Thiele’s dif-
ferential equation can be used for accumulation of the reserve in between
the insurance events. In the Black–Scholes model, we need to distinguish
carefully between the Black–Scholes equation, Equation (4.35), and price
accumulation, Equation (4.36). This is because the stochastic element S
1
t
is
changing continuously.
4.5 Continuous-time insurance model
In this section, we combine the ideas in Section 4.2 with the Black–Scholes
model. More specifically, we consider the continuous-time valuation of the
special claims present in a life insurance contract with bonus. We start by
modifying the main example in Chapter 2, such that it is a purely financial
product and such that investments take place in a Black–Scholes market.
132 Bonus, binomial and Black–Scholes
Then the setup becomes a continuous-time version of the discrete-time setup
in Section 4.2, with investment in the Black–Scholes market instead of in the
binomial market.
The technical reserve V

relates to the first order and the second order
bases, which are made up by the first and second order interest rates r

and
r
o
, through the following differential equation characterizing the technical
reserve:
d
dt
V

(t) =r
o
(t) V

(t) +r
=r

(t) V

(t) +r+o(t) . (4.40)
V

(0) =0.
where the rate of dividends o(t) is given by o(t) =

r
o
(t) −r

(t)

V

(t).
We also speak of the second order interest rate and a process of second order
interest rates

r
o
(t)

0≤t≤n
as the bonus interest rate and the bonus strategy,
respectively.
The pension sum l (t) guaranteed at time t is determined, for a given
technical reserve, by the equivalence relation,
V

(t) =l (t) (ì

)
n−t
−ro

n−t
.
In Chapter 2 we introduced the total reserve U, which appears from an
accumulation under the real basis as follows:
dU (t) =rU (t) dt +rdt.
U (0) =0.
where r is constant. In particular, we know the return over a small time
interval. The idea is to generalize this construction such that we have the
opportunity to invest a part of the reserve in an asset with risk, and we now
allow for investment in the Black–Scholes market. We let the company hold
h
1
(t) stocks at time t.
Changes in the total reserve U consist of premiums and capital gains.
The capital gains consists of the interest rate return on the part of U (t) that
was not invested in stocks, U (t) −r
1
(t) S
1
(t), and the price change on the
portfolio of stocks. Thus,
dU (t) =r

U (t) −r
1
(t) S
1
(t)

dt +r
1
(t) dS
1
(t) +rdt
=rU (t) dt +r
1
(t) S
1
(t) ((o−r) dt +u dW (t)) +rdt.
U (0) =0.
4.5 Continuous-time insurance model 133
The undistributed reserve X is determined residually by
X(t) =U (t) −V

(t) .
and we obtain
dX(t) = dU (t) −dV

(t)
= rX(t) dt +r
1
(t) S
1
(t) ((o−r) dt +u dW (t))
+(c (t) −o(t)) dt.
where
c (t) =(r (t) −r

(t)) V

(t) .
The idea now, as in the discrete-time model in Section 4.2, is to introduce
prospective quantities for the total and the undistributed reserves, respec-
tively. If l (n) is a contingent claim such that it is uniquely determined by the
development of the stock market, the market value of the total payments, the
market value of guaranteed payments and the market value of the unguaran-
teed payments, respectively, are, according to Section 4.4, as follows:
V (t) =E
Q
t
|l (n)] v
n−t
−ro
n−t
. (4.41)
V
g
(t) =l (t) v
n−t
−ro
n−t
.
V
b
(t) =

E
Q
|l (n)] −l (t)

v
n−t
.
Here, E
Q
t
denotes the expectation given the development of the financial
market up to time t and under a particular probability measure, under which
the expected return on investment in the stock is r; see Equation (4.37). In
accordance with the valuation (4.41), residual parameters in the bonus strategy
must be determined by the equivalence relation,
V (0) =0.
As in the discrete-time model in Section 4.2, there exist investment strate-
gies r
1
such that X(n) =0. For example, the investment r
1
(t) =0 brings us
directly back to the situation in Chapter 2. We state a sufficient condition on
r
1
, less trivial than r
1
(t) = 0, for the existence of a bonus strategy leading
to X(n) = 0. If this condition is fulfilled, the calculations in Equation (4.5)
leading to V (t) = U (t) hold. In these calculations we use the fact that the
expected return on U (t) and the premiums paid over (t. n] under Q is r; see
Equation (4.37).
A sufficient condition on r for a fulfilment of X(n) = 0 is that risky
investments are made only if U (t) exceeds V
g
(t). Otherwise, investments
134 Bonus, binomial and Black–Scholes
must give the return r in order to meet the obligations on which V
g
(t) is
based. We write the condition as follows:
U (t) −V
g
(t) =0 ⇒r
1
(t) =0.
An example where this condition is fulfilled is given by
r
1
(t) S
1
(t) =0 (U (t) −V
g
(t)) .
for a constant 0 whereby a constant part of the bonus potential is invested in
stocks. We note that we can allow for 0 >1.
If we add the classical solvency rule X(t) ≥0, a sufficient condition on r
is that risky investments are made only if X(t) ≥ 0. We write the condition
as follows:
X(t) =0 ⇒r
1
(t) =0.
An example where this condition is fulfilled is given by
r
1
(t) S
1
(t) =0X(t) .
for a constant 0 whereby a constant part of the collective bonus potential is
invested in stocks. Again, we can allow for 0 >1.
The conditions on X limit the risk in the portfolio. However, one may
wish to disregard these conditions and invest more capital in stocks. One can
imagine a general investment where the proportion of funds invested in risky
assets, r
1
(t) S
1
(t) ¡X(t), is a function O of (t. V

(t) . X(t)), i.e.
r
1
(t) S
1
(t)
X(t)
=O(t. V

(t) . X(t)) . (4.42)
Hereby,
dX(t) =(r dt +O(t. V

(t) . X(t)) ((o−r) dt +u dW (t))) X(t)
+(c (t) −o(t)) dt.
One can think of an example where a constant part of U is invested in
stocks, i.e.
r
1
(t) S
1
(t) =0U (t) ⇔
O(t. V

(t) . X(t)) =
r
1
(t) S
1
(t)
X(t)
=
0U (t)
X(t)
=0
V

(t) +X(t)
X(t)
.
If we disregard the condition X(n) = 0, one has to discuss, as in the
discrete-time insurance model, the question: how is the bonus strategy linked
to the stock market? And again, we suggest an explicit connection which
4.5 Continuous-time insurance model 135
is practically meaningful and mathematically tractable. However, we stress
that this is only one out of many meaningful constructions. We let the bonus
interest rate be linked to X(n) by defining the following:
r
o
(t) =r

(t) +
4(t. X(t))
V

(t)
.
where 4 is a function specifying the dependence on X. One can think of a
function in the form
4(t. x) =(8(t) +o(t) x)
+
. (4.43)
where o and 8 are deterministic functions of t. We now consider valuation
under the three bonus forms, terminal bonus, cash bonus and accumulation of
benefits, with exactly that choice of function 4. We still speak of the function
4 as the contract function and recall the two possible constructions included
in Equation (4.43): if 8(t) =0 and o(t) =o >0, we obtain
4(t. X(t)) =o(X(t))
+
.
i.e. the dividend is determined as a constant part of the undistributed reserve,
if this is positive.
If o(t) =1 and 8(t) =−K -0, we obtain
4(t. X(t)) =(X(t) −K)
+
.
i.e. the dividend is determined such that the part of the undistributed reserve
that exceeds a constant buffer K is paid out.
4.5.1 Terminal bonus
We start by considering the terminal bonus because this is the bonus form
which has most similarities to the European option. This provides us with our
first generalization of the Black–Scholes equation. We let the investment be
determined by a function O, such that r
1
(t) S
1
(t) ¡X(t) =O(t. X(t)). This
is not a restriction compared with Equation (4.42) since the technical reserve
under terminal bonus is a deterministic function of time. Terminal bonus is
given by letting the second order basis coincide with the first order basis
throughout the term of the contract and then, at the terminal time n, paying
out a bonus sum. Hereby, the dynamics for X(t) becomes
dX(t) =(rdt +O(t. X(t)) ((o−r) dt +u dW (t))) X(t) +c (t) dt.
c (t) =(r (t) −r

(t)) V

(t) .
136 Bonus, binomial and Black–Scholes
i.e. the full safety margin makes up the saving rate and is added to the
undistributed reserve.
No dividends are added to the technical reserve in the contract period, so we
have l (t) =l (0). On the other hand, at termination the company pays out, in
addition to l (0), a bonus sum, which we assume can be written as 4(X(n))
for a deterministic contract function of the form given in Equation (4.43).
Note that 4 plays a different role here than in the preceding sections because
we now consider a different bonus form. It is now possible to derive, based on
arbitrage arguments, a deterministic differential equation for the total reserve
V which generalizes the Black–Scholes equation:
o
ot
V (t. x) = rV (t. x) +r

o
ox
V (t. x) (rx+c (t))

1
2
O
2
(t. x) u
2
x
2
o
2
ox
2
V (t. x) . (4.44)
V (n. x) = l (0) +4(x) .
and we can now relate the different terms to the elements of Thiele’s dif-
ferential equation and the Black–Scholes equation. The first line contains
the risk-free return on the value and the rate of premiums recognized from
Thiele’s differential equation. The second line contains the partial derivative
with respect to x. This corresponds to the partial derivative with respect to s
in the Black–Scholes equation but the factor is modified such that it equals
the drift coefficient in the dynamics of X. Here, we remember to replace o
by r. The third line contains the second order derivative with respect to x and
corresponds to the second order derivative with respect to s in the Black–
Scholes equation. The factor equals half of the squared coefficient in front of
dW (t) in the dynamics of X. We refer to Steffensen (2000) for details and
for the derivation of Equation (4.44).
The differential equation (4.44) is a deterministic tool for the determination
of V (t. X(t)). A differential equation for accumulating the total reserve, cor-
responding to the price dynamics, Equation (4.36), can be derived, however,
as follows:
dV (t. X(t)) = rV (t. X(t)) dt +rdt
+
o
ox
V (t. X(t)) O(t. X(t)) X(t) ((o−r) dt +u dW (t)) .
4.5 Continuous-time insurance model 137
Comparing this with the dynamics in Equation (4.23), it is possible, taking
into consideration the premium r, to construct a hedging portfolio from the
following relation which determines h
1
:
o
ox
V (t. X(t)) O(t. X(t)) X(t) =h
1
(t) S
1
(t) .
and h
0
is then determined residually.
The solution of Equation (4.44) has the following representation formula:
V (t. u) =e
−r(n−t)

l (0) +E
Q
t.x
|4(X(n))]

n
t
e
−r(s−t)
rds.
where E
Q
t.x
denotes the expectation under Q conditioned on X(t) = x. For
certain specifications of O and 4, it is possible to derive an explicit formula.
Prudent investment
As an example of an explicit formula, we show what happens if we invest
so prudently that X(n) ≥ 0. Then we can assume a contract function in the
form 4(x) = 8+ox, 8 ≥ 0, o ≥ 0, and we can now “guess” a solution of
the following form:
V (t. x) =1 (t) +g (t) x.
Substituting this solution into Equation (4.44), we find that V (t. x) fulfils
Equation (4.44) if 1 and g fulfil the following differential equations:
1

(t) =r1 (t) +r−gc (t) .
g

(t) =0.
Together with the terminal conditions 1 (n) = l (0) +8 and g (n) = o, we
obtain the solution to that system as follows:
g (t) =o.
1 (t) =

n
t
e
−r(s−t)
(oc (s) −r) ds +e
−r(n−t)
(l (0) +8) .
We now wish to find a fair combination of the parameters in the contract
function, o and 8; this is achieved by the fulfilment of the equivalence relation
V (0. 0) =0, i.e. 1 (0) =0, where
1 (0) =

n
0
e
−rs
(oc (s) −r) ds +e
−rn
(l (0) +8) .
138 Bonus, binomial and Black–Scholes
It is now easy to see that one fair construction (out of many) is given by
(o. 8) =(1. 0), corresponding to paying out the full undistributed reserve as
bonus at time n. Thereby,
g (t) =1.
1 (t) =

n
t
e
−r(s−t)
(c (s) −r) ds +e
−r(n−t)
l (0) .
and we see how V (t) = 1 (t) +X(t) consists of X(t) plus the individ-
ual bonus potential,

n
t
exp(−r (s −t)) c (s) ds, plus the market value of the
guaranteed payments, e
−r(n−t)
l (0) −

n
t
exp(−r (s −t)) rds. One may argue
that the special case (o. 8) = (1. 0) is uninteresting since then the surplus
X appropriately defined at time n, after subtraction of the terminal bonus,
equals zero. So, in this case, we actually have that V (t) = U (t), and there
is no reason to perform the calculations above. However, the derivation still
serves as a demonstration of techniques.
4.5.2 Cash bonus
We now consider the bonus form cash bonus. We let the investment be
determined by a function O, such that r
1
(t) S
1
(t) ¡X(t) = O(t. X(t)). As
was the case for terminal bonus, this is not a restriction compared with
Equation (4.42) since the technical reserve is a deterministic function of t.
Cash bonus is given by paying out the dividends, according to a second order
basis, immediately to the policy holder. Thus, guaranteed payments r and
l (0) fall due, but in addition the policy holder receives a cash bonus payment
of

r
o
(t) −r

(t)

V

(t). The policy holder can view this as a discount on
the premiums. We can link the bonus payments to the undistributed reserve
through the contract function 4 by letting
r
o
(t) =r

(t) +
4(t. X(t))
V

(t)
.
Thereby, the bonus payments and the dynamics for X(t) become

r
o
(t) −r

(t)

V

(t) =4(t. X(t))
and
dX(t) = (r dt +O(t. X(t)) ((o−r) dt +u dW (t))) X(t)
+c (t) dt −4(t. X(t)) dt.
4.5 Continuous-time insurance model 139
respectively. One can think of a contract function in the form of Equation (4.43).
Again, it is possible to derive a differential equation for the function V (t. x),
which becomes
o
ot
V (t. x) = rV (t. x) +r−4(t. x)

o
ox
V (t. x) (rx+c (t) −4(t. x))

1
2
O
2
(t. x) u
2
x
2
o
2
ox
2
V (t. x) . (4.45)
V (n. x) = l (0) .
and the elements can be related to those of Thiele’s differential equation and
the Black–Scholes equation in basically the same way as in Section 4.5.1.
Again, we refer to Steffensen (2000) for details and for the derivation of
Equation (4.45).
The differential equation for accumulating the total reserve becomes
dV (t. X(t)) = rV (t. X(t)) dt +rdt −4(t. X(t)) dt
+
o
ox
V (t. X(t)) (O(t. X(t)) ((o−r) dt +u dW (t)) X(t)) .
and the hedging portfolio of stocks is, according to (4.22) and taking into
consideration the premium payments and cash bonus, determined by
o
ox
V (t. X(t)) O(t. X(t)) X(t) =h
1
(t) S
1
(t) .
The solution to Equation (4.45) has the following stochastic representation
formula:
V (t. x) =e
−r(n−t)
l (0) +

n
t
e
−r(s−t)

E
Q
t.x
|4(s. X(s))] −r

ds.
For certain specifications of O and 4, it is possible to find an explicit formula.
Prudent investment
As an example of an explicit formula, we show what happens if we invest
so prudently that X(t) ≥ 0. Then we can assume a contract function in the
form 4(t. x) =8(t) +o(t) x, o(t) ≥ 0, 8(t) ≥ 0, and we can now “guess”
a solution of the following form:
V (t. x) =1 (t) +g (t) x.
140 Bonus, binomial and Black–Scholes
Substituting this solution into Equation (4.45), we find that V (t. x) fulfils
(4.45) if 1 and g fulfil the following differential equations:
1

(t) =r1 (t) +r−8(t) −g (t) c (t) +8(t) g (t) .
g

(t) =rg (t) −o(t) +g (t) o(t) .
Together with the terminal conditions 1 (n) =l (0) and g (n) =0, we obtain
the solution to that system as follows:
g (t) =

n
t
e

s
t
o(t)dt
o(s) ds.
1 (t) =

n
t
e
−r(s−t)
(8(s) −r+g (s) (c (s) −8(s))) ds +e
−r(n−t)
l (0) .
Now we find a fair contract by determining (o. 8) such that the equivalence
relation V (0. 0) =0 is fulfilled, i.e. 1 (0) =0.
It is now easy to see that one fair construction (out of many) is given
by (o. 8) = (0. c), corresponding to paying out the contributions as bonus.
Thereby,
g (t) =0.
1 (t) =

n
t
e
−r(s−t)
(c (s) −r) ds +e
−r(n−t)
l (0) .
and we see how V (t) =1 (t) consists of the individual bonus potential given
by

n
t
exp(−r (s −t)) c (s) ds plus the market value of the guaranteed pay-
ments given by e
−r(n−t)
l (0)−

n
t
exp(−r (s −t)) rds. One may argue that the
special case (o. 8) =(0. c) is uninteresting since then X(t) =0 for all t and,
in particular, X(n) =0. So, in this case, we actually have that V (t) =U (t),
and there is no reason to perform the calculations above. However, the deriva-
tion still serves as a demonstration of techniques.
4.5.3 Additional benefits
Finally, we consider the bonus form accumulation of benefits. This is the
bonus form described in Chapter 2 where the dividend payments are added
to the technical reserve and where the guaranteed pension sum is increased
correspondingly. Thus, the guaranteed payments r and l (n) fall due, where
l (n) = V

(n) is the terminal pension sum, including accumulation of past
dividends. We let the investment be determined by a function O such
that r
1
(t) S
1
(t) ¡X(t) = O(t. V

(t) . X(t)), and again we link the bonus
4.5 Continuous-time insurance model 141
payments to the undistributed reserve through a contract function 4, i.e. by
r
o
(t) =r

(t) +
4(t. V

(t) . X(t))
V

(t)
.
Thereby, the dividend payments and the dynamics for X and V

become

r
o
(t) −r

(t)

V

(t) =4(t. X(t)) .
dX(t) = (r dt +O(t. V

(t) . X(t)) ((o−r) dt +udW (t))) X(t)
+(r (t) −r

(t)) V

(t) dt −4(t. X(t) . V

(t)) dt
and
dV

(t) =r

V

(t) dt +rdt +4(t. X(t) . V

(t)) dt.
respectively. One can think of a contract function in the form of Equa-
tion (4.43). It is now possible to derive a differential equation for the function
V (t. v. x), which becomes
o
ot
V (t. v. x) = rV (t. v. x) +r

o
ox
V (t. v. x) (rx+(r −r

) v −4(t. v. x))

o
ov
V (t. v. x) (r

v +r+4(t. v. x))

1
2
O
2
(t. v. x) u
2
x
2
o
2
ox
2
V (t. v. x) . (4.46)
V (n. v. x) = v.
The elements can be related to Thiele’s differential equation and the Black–
Scholes equation in basically the same way as in Section 4.5.1. Again, we
refer to Steffensen (2000) for details and for the derivation of Equation (4.46).
The differential equation for accumulating the total reserve becomes
dV (t. V

(t) . X(t)) =rV (t. V

(t) . X(t)) dt +rdt
+
o
ox
V (t. V

(t) . X(t)) O(t. V

(t) . X(t)) X(t)
×((o−r) dt + udW (t)) .
and the hedging portfolio of stocks is, according to Equation (4.22) and taking
into consideration the premiums, determined by
o
ox
V (t. V

(t) . X(t)) O(t. V

(t) . X(t)) X(t) =h
1
(t) S
1
(t) .
142 Bonus, binomial and Black–Scholes
The solution to Equation (4.46) has the following stochastic representation
formula:
V (t. v. x) =

n
t
e
−r(s−t)

E
Q
t.v.u
|V

(n)] −r

ds.
where E
Q
t.v.u
denotes the expectation under Q conditioned on V

(t) = v and
X(t) = x. For certain specifications of O and 4, it is possible to find an
explicit formula.
Prudent investment
As an example of an explicit formula, we show what happens if we invest so
prudently that X(t) ≥0. Then we can assume a contract function in the form
4(t. v. x) = 8(t) +o(t) x, o(t) ≥ 0, 8(t) ≥ 0, and we can now “guess” a
solution of the following form:
V (t. v. x) =1 (t) +g (t) x+h(t) v.
Substituting this solution into Equation (4.46), we find that V (t. v. x) fulfils
(4.46) if 1, g and h fulfil the following differential equations:
1

(t) =r1 (t) +r+(g (t) −h(t)) 8(t) −h(t) r.
g

(t) =g (t) o(t) −h(t) o(t) .
h

(t) =(r −r

) h(t) −g (t) (r −r

) .
Together with the terminal conditions 1 (n) =0, g (n) =0, and h(n) =1, the
solution to that system becomes
g (t) =

n
t
e

s
t
o
o(s) h(s) ds.
h(t) =e
−(r−r

)(n−t)
+

n
t
e
−(r−r

)(s−t)
(r −r

) g (s) ds.
1 (t) =

n
t
e
−r(s−t)
(h(s) (r+8(s)) −g (s) 8(s) −r) ds.
The solution to this system can also be written using so-called matrix expo-
nentials. Now we find a fair contract by determining (o. 8) such that the
equivalence relation V (0. 0. 0) =0 is fulfilled, i.e. 1 (0) =0.
It is now easy to see that one fair construction (out of many) is given by
(o. 8) = (0. c), corresponding to paying out the contributions as dividends.
Thereby,
4.6 Generalizations of the models 143
g (t) =0.
h(t) =e
−(r−r

)(n−t)
.
1 (t) =

n
t
e
−r(s−t)
(h(s) c (s) −r(1−h(s))) ds.
and we see how V (t) = 1 (t) +h(t) V

(t) consists of two parts. One part,
h(t) V

(t), is actually the market value of the free policy benefits. By dif-
ferentiation of V with respect to t, one sees that actually V (t) = U (t),
and from this we conclude that the other part of V in the representation
V (t) = 1 (t) +h(t) V

(t), 1 (t), expresses the bonus potential on the free
policy, i.e. 1 (t) = V
bf
(t); see Chapter 2. One may argue that the special
case (o. 8) =(0. c) is uninteresting since V (t) =U (t) and there is no reason
to perform the calculations above. However, the derivation still serves as a
demonstration of techniques.
4.6 Generalizations of the models
In this chapter the financial market has been modeled by the binomial
model and the Black–Scholes model. Often, these models are too simple
to cover the effects experienced from the real financial market. In partic-
ular, when evaluating life insurance obligations with terms of more than
thirty years, these models are far too simple. In this final section we there-
fore consider some possible generalizations which make the models more
realistic.
Stochastic interest in the binomial model
A binomial model for the interest rate was introduced in Chapter 3. It is
possible to combine the binomial model introduced in Sections 4.2 and 4.3
with a binomial model for the interest rate. We introduce a sequence of two-
dimensional stochastic variables

Z
r
(t) . Z
S
(t)

t=1.... .n
, where each of the
variables Z
r
(t) and Z
S
(t) assume one of two values and drives the stochastic
interest rate and the stock, respectively. Note that Z
r
(t) and Z
S
(t) may be
correlated. One can now put up a pricing formula for the conditional claim
X falling due at time n in correspondence with the principle of no arbitrage.
Unfortunately, one does not obtain a unique formula but a continuum of
formulas, as follows:
H
Q
(t. X) =E
Q
t
¸
1+r (t)
1+r (n)
X
¸
. Q∈ Q. (4.47)
144 Bonus, binomial and Black–Scholes
where Q is now a set of martingale measures which all give arbitrage-free
prices. All martingale measures, i.e. measures under which the discounted
stock price,
1
1+r (1)
· · ·
1
1+r (t)
S (t) .
is a martingale, can be used here.
It is indeed a drawback, compared with the binomial model with determin-
istic interest rate, that we do not obtain unique pricing formulas. However, if
we introduce a bond market with observable bond prices, we may again obtain
unique prices. This corresponds to pointing out the unique martingale mea-
sure under which not only the discounted stock prices, but also the discounted
bond prices are martingales. This relates to the situation in Chapter 3, where
Q is determined uniquely after the introduction of observable zero coupon
bond prices P
n
t
. If these are known, and Z
r
and Z
S
are independent under the
measure Q, Equation (4.47) reduces to
H(t. X) =P (t. n) E
Q
t
|X] . (4.48)
Stochastic interest in the Black–Scholes model
As in the binomial model, it would be natural to allow for a stochastic interest
rate in the Black–Scholes model. In Chapter 3 a stochastic model is introduced
where the dynamics of the interest rate are given by
dr (t) =ì (t. r (t)) dt +u (t. r (t)) dW
r
(t) .
where W
r
is a Wiener process. We can now replace the deterministic interest
rate in the Black–Scholes model by such an interest rate. Equivalent to the
situation in discrete time where the binomial stochastic variables determining
the interest rate and the stock could be correlated, we can also let the two
Wiener processes, W and W
r
, be correlated. The set of arbitrage-free prices
of a contingent claim X becomes
H
Q
(t. X) =E
Q
t
¸
e

n
t
r(s)ds
X
¸
. Q∈ Q. (4.49)
where Q is the set of martingale measures. As was discussed in Chapter 3, we
require a knowledge of the prices at the bond market to determine a unique
martingale measure. If we know the zero coupon prices P (t. n) and assume
independence between r and S
1
under Q, we obtain the pricing formula given
by Equation (4.48) again.
4.6 Generalizations of the models 145
Factor models
The stochastic interest rate process introduced above in a Black–Scholes
model belongs to a set of generalized financial models which are called
factor models. A K-factor model for a continuous financial market is a model
consisting of a K-dimensional factor Y =

Y
1
. . . . . Y
K

, where the dynamics
of Y
k
is given by
dY
k
(t) =µ
k
(t. Y (t)) dt +
n
¸
i=1
u
ki
(t. Y (t)) dW
i
(t) .
where W =

W
1
. . . . . W
n

is an n-dimensional Wiener process. These factors
determine a risk-free interest rate r (t. Y (t)). We are interested in prices and
hedging portfolios connected to conditional claims in the form 4(Y), where
4 is the contract function. If “a lot” of the factors are prices on traded assets,
then we can say “a lot” about prices and hedging portfolios. If only “a few”
of the factors are prices on traded assets, then we can say only “a little” about
prices and hedging portfolios.
The example with stochastic interest rate in the Black–Scholes model
above is a two-factor model where the interest rate itself plays the role of
factor. Another popular class of factor models introduces stochastic processes
for the volatility u = (u (t))
t≥0
in the Black–Scholes market, the stochastic
volatility models. In Chapter 3, a Cox–Ingersoll–Ross process was mentioned
as a candidate for the stochastic interest rate. This process is suggested by
Heston (1993) for the description of the square of u (t), i.e. ~ (t) = u
2
(t),
with the dynamics given by
d~ (t) =o(l −~ (t)) dt +c

~ (t) dW
~
(t) .
where W
~
is a Wiener process driving the volatility.
In general, we cannot obtain unique arbitrage-free prices in factor models.
This relates to the fact that we cannot determine a unique martingale measure,
which again relates to the fact that such models are not, in general, complete.
Another important example of an incomplete market appears by introducing
life insurance obligations under the consideration of mortality risk which is
not to be diversified, such as, For example, by introduction of stochastic
mortality intensities. Parts of Møller (2000) and Steffensen (2001) deal with
questions arising when dealing with incomplete markets. Chapter 5 gives an
introduction to some useful tools in incomplete markets.
5
Integrated actuarial and financial valuation
5.1 Introduction
In the previous chapters it has been assumed that mortality risk is
diversifiable. Under this assumption, we have proposed valuation principles
which could be derived by working with a big portfolio of insured lives
(allowing for deterministic decrement series) or by letting the size of the
portfolio converge to infinity. In Chapter 2 the financial market consisted of
one single investment possibility, a risk-free asset corresponding to a savings
account with a deterministic interest. Chapter 3 considered the case where
the interest rate is stochastic and derived formulas for the market value of the
guaranteed payments that involved the prices of zero coupon bonds. Finally,
Chapter 4 studied a financial market with two investment possibilities, a stock
and a savings account, and demonstrated how the market value of the total
liabilities (including bonus) could be determined for fixed investment and
bonus strategies. This study led to explicit formulas for the market value
of the total liabilities in two specific models: the binomial model and the
Black–Scholes model under the assumption of no arbitrage and diversifiable
mortality risk. In this case, the market value was defined as the amount which
was necessary in order to hedge the liabilities perfectly via a self-financing
investment strategy.
The main goal of the present chapter is to analyze the combined, or
integrated, insurance and financial risk which is present in a life insurance
contract, where the benefits are linked to returns on the financial markets.
For with-profit insurance contracts, one can model the future investment and
bonus strategies of the company in order to obtain a description of the link
between future benefits and the development of the traded assets using the
setup from Chapter 4. Here, we focus on unit-linked life insurance contracts,
where benefits are typically already linked to the returns on certain assets in
146
5.1 Introduction 147
a specific way, which is stipulated in the insurance contract. Thus, with a unit-
linked insurance contract, it is in principle the policy holder who decides
where the premiums should be invested and how the benefits should be linked
to the actual return on these investments; for example, the policy holder can
decide whether or not the contract should include minimum guarantees. Thus,
in principle, the contract specifies completely how the benefits and the policy
holder’s account develops in any financial scenario.
Let us now consider the liabilities from a given portfolio of life insur-
ance contracts as an integrated risk or integrated contingent claim, i.e. as
a contingent claim which depends on some financial risk as well as some
insurance risk. In this case, it is no longer possible to derive unique market
values for the liability from the assumption of no arbitrage alone. Since we
typically cannot trade with mortality in the financial markets, these markets
cannot tell how we should price and hedge mortality risk. The consequence
of this observation is that we cannot find any self-financing strategies that
hedge the liabilities perfectly. For example, how would you hedge a contract
which pays one million euros at a future time T, provided that a certain policy
holder is alive at this time, and zero otherwise, by trading in a stock and a
savings account which evolves independently of the policy holder’s lifetime?
One possible way of solving this problem is to consider models where there
exist traded assets which determine the market’s attitude towards mortality
risk. For more details on this approach, see Møller (1998) and Steffensen
(2001).
An alternative way of solving this problem could be to charge a premium,
which, provided that it is invested in the savings account, is sufficient to
pay the one million euros whether the policy holder survives or not. One
problem with this approach is of course that it essentially corresponds to
using a survival probability of 1, and hence completely neglects the element
of insurance involved. Another problem is that this method does not replicate
the liability completely, since it leads to a surplus of one million euros if
the policy holder is not alive at time T. Investment strategies which ensure
sufficient (but possibly too much) capital in order to cover the liabilities are
also called super-hedging strategies. The present chapter demonstrates that
super-hedging in general leads to unreasonably high prices for life insurance
contracts. In addition, we treat various other principles for the pricing and
hedging of the total liability. These principles have been developed for so-
called incomplete markets, that is for markets where there are contingent
claims that cannot be hedged perfectly by using a self-financing strategy.
The principles are based on some relatively advanced mathematics, which
are not covered here; however, some technical details can be found in the
148 Integrated actuarial and financial valuation
Appendix. The chapter focuses instead on the basic ideas and tries to illustrate
the consequences of applying these advanced methods to a portfolio of unit-
linked or with-profit life insurance contracts.
The chapter is organized as follows. Section 5.2 gives an introduction
to unit-linked life insurance and demonstrates how the liabilities may be
viewed as a contingent claim. Section 5.3 proposes a simple system for the
modeling of the policy holder’s account under unit-linked insurance contracts
with guarantees, and Section 5.4 repeats the usual argument for pricing under
diversification for a large portfolio. In Section 5.5, we bring to the surface
the combined insurance and financial risk inherent in a unit-linked insurance
contract by working within a one-period model similar to the one considered
in Chapter 4. We then derive strategies that minimize the variance on the
company’s total costs, defined as the company’s liability reduced by premiums
and investment gains. In addition, we give examples of models where the
mortality risk cannot be eliminated by increasing the size of the portfolio.
Section 5.6 is devoted to a review of the multi-period model, which was
treated in Chapters 3 and 4, and in Section 5.7 we analyze the combined risk
in unit-linked contracts within the multi-period model. We describe principles
for valuation that take into account the risk that can be hedged in the financial
market as well as the risk that cannot be controlled by trading in these markets.
In Section 5.8, comparisons are made between unit-linked and traditional
with-profit life insurance contracts. Parts of the present chapter follow Møller
(2001a, 2002).
5.2 Unit-linked insurance
A unit-linked life insurance contract differs from traditional life insurance
contracts in that benefits (and possibly also premiums) are linked directly to
the value of a unit of some investment portfolio. Other names used for these
contracts are equity-linked, equity-based or variable-life (in the USA). This
construction allows for a high degree of flexibility in that the policy holder
in principle can affect the way premiums are invested. In this way, one can
adapt the investment strategy to the policy holder’s needs and preferences.
One obvious idea is to apply a more aggressive investment strategy for young
policy holders than the one used for policy holders which are close to the
age of retirement. In addition, it is possible to offer contracts that are linked
to certain stock indices, for example world indices, country-specific indices
or reference portfolios with more or less clearly defined investment profiles;
5.2 Unit-linked insurance 149
examples are stocks from certain lines of businesses, areas or companies with
specific ethical codes. Unit-linked insurance contracts seem to have been
introduced for the first time in Holland in the beginning of the 1950s. In
the USA the first unit-linked insurance contracts were offered around 1954,
and in the UK the first contracts appeared in 1957. An overview of the early
development of these contracts can be found in Turner (1971).
In this chapter we discuss the development of the policy holder’s account
and risk management for some of the unit-linked contracts that are offered
in practice. However, we first treat some more basic aspects in a simplified
framework. In particular, we view portfolios of unit-linked life insurance
contracts as integrated risks, which include financial and insurance risk. The
financial risk is related to the development of the underlying assets, whereas
the insurance risk is related to the uncertain number of survivors. Within this
framework, one can ask questions such as: when is it possible to hedge this
integrated risk? How big should the portfolio be in order to guarantee that
the mortality risk is diversifiable? How can the total risk inherent in a life
insurance portfolio be characterized if mortality risk is no longer assumed to
be diversifiable?
5.2.1 Unit-linked life insurance as a contingent claim
Consider an insurance company with a portfolio of unit-linked life insurance
contracts that are linked to the value of some asset, a stock index or an
investment portfolio (henceforth simply called the index). Denote by S
1
(t)
the value at time t of this index. In the following, we refer to the entire future
development of the value of the index until the time of maturity T by simply
writing S
1
; in principle, we should use the notation (S
1
(t))
t∈|0.T ]
in order to
underline the fact that we are working with the entire price process. In our
example, we consider an idealized portfolio which consists of /
x
policy holders
with identically distributed lifetimes T
1
. . . . . T
/
x
, who have all purchased the
same unit-linked pure endowment paid by a single premium r(0) at time 0.
We simplify further by assuming that the lifetimes are independent. With the
contract considered, the amount 1(S
1
) is payable upon survival to time T. One
can think of 1 as a function which describes exactly how the sum insured,
payable upon survival to T, depends explicitly on the development in the
value of the underlying index. We first focus on the problem of valuating the
benefits, and hence we do not explicitly deal with relations between benefits
and premiums.
150 Integrated actuarial and financial valuation
The present value at time 0 of the company’s liabilities is calculated by
discounting the future payments, i.e.
H =
/
x
¸
i=1
1
]T
i
>T]
1(S
1
)e

T
0
r(u)du
. (5.1)
where we have used the market interest rate r as described in Chapter 3. We
can think of Equation (5.1) as a contingent claim, which depends on both the
value of the stock index and the number of survivors.
Simple examples are obtained by allowing the amount payable upon sur-
vival to T to depend on the terminal value of the stock index only, for
example
1(S
1
) =S
1
(T). (5.2)
or the terminal value guaranteed against falling short of some fixed amount K,
1(S
1
) =max(S
1
(T). K). (5.3)
The contract with Equation (5.2) is also known as a pure unit-linked con-
tract, and Equation (5.3) is a unit-linked contract with terminal guarantee
(the guaranteed benefit is in this case K). However, one can also construct
more complicated functions of the process S
1
, for example by considering a
contract with yearly guarantees with
1(S
1
) =K·
T
¸
t=1
max

1+
S
1
(t) −S
1
(t −1)
S
1
(t −1)
. 1+o(t)

. (5.4)
Here, the quantity (S
1
(t) −S
1
(t −1))¡S
1
(t −1) is the relative return from the
index in year t and o(t) is the guaranteed return in year t. At time 0, the
benefit is guaranteed against falling short of

T
¸
t=1
(1+o(t)).
but the guarantee is increased during the term of the contract if the yearly
relative returns exceed the yearly guaranteed returns.
In other situations, it is relevant to study payment functions of the following
form:
1(S
1
) =max

T−1
¸
t=0
S
1
(T)
S
1
(t)
.
T−1
¸
t=0

T
¸
s=t+1
(1+o(s))

.
which arise in the case of periodic premiums, where the policy holder invests
one unit at times t =0. 1. . . . . T −1. At time t, this leads to 1¡S
1
(t) units of
the underlying index, and the value of this investment changes to S
1
(T)¡S
1
(t)
5.2 Unit-linked insurance 151
at time T. The benefit at T is now calculated as the maximum of the total
value at time T of the premiums invested in the index and the value if all
premiums had been invested in an account with periodic return o(t).
Unit-linked contracts have been analyzed by actuaries and others since the
late 1960s; see, for example, Kahn (1971), Turner (1969) and Wilkie (1978).
Kahn (1971) and Wilkie (1978) give simulation studies for an insurance
company administrating portfolios of unit-linked insurance contracts. Using
modern theories of financial mathematics, Brennan and Schwartz (1979a, b)
have suggested new valuation principles and investment strategies for unit-
linked insurance contracts with so-called asset value guarantees (minimum
guarantees). Their principles essentially consisted in combining traditional
(law of large numbers) arguments from life insurance with the methods of
Black and Scholes (1973) and Merton (1973). By appealing to the law of
large numbers, Brennan and Schwartz (1979a, b) first replaced the uncertain
courses of the insured lives by their expected values (see also the argument
given in Chapter 3, in particular Section 3.2.4). Thus, the actual insurance
claims including mortality risk as well as financial risk were replaced by
modified claims, which only contained financial uncertainty. More precisely,
instead of considering the claim in Equation (5.1), they looked at
H

=/
x T
¡
x
1(S
1
)e

T
0
r(u)du
=/
x+T
1(S
1
)e

T
0
r(u)du
. (5.5)
where we have used standard actuarial notation
T
¡
x
=P(T
1
>T) and /
x+T
=
/
x T
¡
x
. This modified liability could then essentially be identified with an
option (albeit with a very long maturity), which could in principle be priced
and hedged using the basic principles of (modern) financial mathematics
as described in Chapter 4. (However, the time to maturity of an option is
typically less than one year, whereas life insurance contracts often extend to
more than fifteen years.) For the pure unit-linked contract in Equation (5.2),
i.e. the contract without a guarantee, the liability (5.5) is proportional to the
discounted terminal value of the stock S
1
(T) at time T. This liability can
be hedged by a so-called buy-and-hold strategy, which consists in buying
/
x T
¡
x
units of the stock at time 0 and holding these until T. Thus, in the
case of no guarantee, the usual arbitrage argument shows that the unique
arbitrage-free price of H

is simply /
x+T
S
1
(0). Consequently, one possible
fair premium for each policy holder is
T
¡
x
S
1
(0), the probability of survival to
T times the value at time 0 of the stock index. Now consider the contract with
benefit 1(S
1
) =max(S
1
(T). K) =(S
1
(T) −K)
+
+K. In this case, the pricing
of Equation (5.5) involves the pricing of a European call option, since
H

=/
x+T
e

T
0
r(u)du
K+/
x+T
e

T
0
r(u)du
(S
1
(T) −K)
+
.
152 Integrated actuarial and financial valuation
In the general case, we see that this principle suggests the premium
/
x T
¡
x
V(0. 1) =/
x+T
V(0. 1).
where V(0. 1) is the price at time 0 of the purely financial contract which
pays 1(S
1
) at time T.
More recently, the problem of pricing unit-linked life insurance contracts
(under constant interest rates) has been addressed by Aase and Persson (1994),
Bacinello and Ortu (1993a) and Delbaen (1986), among others, who com-
bined the so-called martingale approach of Harrison and Kreps (1979) and
Harrison and Pliska (1981) with law of large numbers arguments. Bacinello
and Ortu (1993b), Bacinello and Persson (1998) and Nielsen and Sandmann
(1995), among others, generalized existing results to the case of stochastic
interest rates. For a treatment of computational aspects and design of unit-link
contracts, see Hardy (2003).
Aase and Persson (1994) worked with continuous survival probabilities
(i.e. with death benefits that are payable immediately upon the death of the
policy holder and not at the end of the year as would be implied by discrete-
time survival probabilities) and suggested investment strategies for unit-linked
insurance contracts using methods similar to the ones proposed by Brennan
and Schwartz (1979a, b) for discrete-time survival probabilities. Whereas
Brennan and Schwartz (1979a, b) considered a large portfolio of policy holders
and therefore worked with deterministic mortality, Aase and Persson (1994)
considered a portfolio consisting of one policy holder only. However, since
the random lifetime had already been replaced by the expected course in
order to allow for an application of standard financial valuation techniques
for complete markets, the resulting strategies were not able to account for the
mortality uncertainty within a portfolio of unit-linked life insurance contracts.
This leaves open the question of how to quantify and manage the combined
actuarial and financial risk inherent in these contracts. In particular, one can
investigate to what extent it is possible to hedge liabilities by trading in the
financial market.
5.3 The policy holder’s account
This section describes the development (in discrete time) of the policy holder’s
account associated with a unit-linked insurance contract.
Consider a premium payment plan r(0). r(1). . . . . r(T −1) (gross premi-
ums), and define the corresponding net premiums by (1−~(t))r(t). Denote
by V

(t) the value at time t of the policy holder’s account after payment
5.3 The policy holder’s account 153
of the net premium (1 −~(t))r(t). The premiums are currently invested in
an underlying stock index (or a portfolio) whose value at time t is given by
S
1
(t). The relative return a(t) on this index in the period (t −1. t] is given by
a(t) =
S
1
(t) −S
1
(t −1)
S
1
(t −1)
.
Our framework allows for contracts with guarantees, which may be changed
during the term of the contract as a result of new premiums or as a consequence
of the development on the financial markets. The guarantee can be specified
in terms of the yearly returns a(t) on the index, and hence it may affect not
only the sum insured, but also the current value of the account. Alternatively,
the guarantee can be formulated directly in terms of the amounts payable at
the time of maturity of the contract (at time T). Denote by l
ad
(t) the sum
payable at time t in case of a death during the interval (t −1. t]. The basic
idea is that the policy holder receives the value V

(T) of the account upon
survival to T. We consider situations where the payment upon survival is
guaranteed against falling short of some minimum guarantee G(T).
The development of the policy holder’s account during some time interval
is in general given as follows:
account, beginning of period + investment returns + premiums
− expenses −payment for guarantee
− risk premiums =account, end of period.
With the notation introduced above, the value of the accounts at time 0 for
the portfolio consisting of /
x
policy holders is given by
V
∗port
(0) =/
x
V

(0) =/
x
(1−~(0))r(0).
and the development in the value is taken to be described by the following
recursive formula:
V
∗port
(t) = (1+¯a(t)) V
∗port
(t −1) +/
x+t
(1−~(t))r(t)
−/
x+t−1
ì(t) −J
x+t−1
l
ad
(t). (5.6)
where J
x+t−1
=/
x+t−1
−/
x+t
is the expected number of deaths in the portfolio
between age x +t −1 and x +t. In the recursion, Equation (5.6), ì(t) is
the price of annual or terminal guarantees and ¯a(t) is the return which is
actually credited to the policy holders’ accounts. The term with the premiums
(1 −~(t))r(t) involves the current number of survivors at time t, which
is given by /
x+t
. This means that premiums are paid at time t by each
of the survivors. In contrast, the term with ì(t) involves the number of
154 Integrated actuarial and financial valuation
survivors at time t −1, which indicates that each of the survivors at the
beginning of the period (at time t −1) will be charged at time t for the
guarantee.
If we divide V
∗port
(t) by the current number of survivors at time t, /
x+t
,
we can derive a similar recursion for the development of the account V

(t) =
V
∗port
(t)¡/
x+t
for one policy holder. To see this, note the following:
V

(t) =
V
∗port
(t)
/
x+t
=(1+¯a(t))
/
x+t−1
/
x+t
V
∗port
(t −1)
/
x+t−1
+(1−~(t))r(t)

/
x+t−1
/
x+t
ì(t) −
J
x+t−1
/
x+t
l
ad
(t)
=(1+¯a(t))V

(t −1) +(1−~(t))r(t) −
/
x+t−1
/
x+t
ì(t)
+(1+¯a(t))

/
x+t−1
/
x+t
−1

V

(t −1) −
J
x+t−1
/
x+t
l
ad
(t). (5.7)
where the second equality follows by inserting Equation (5.6), and where the
third equality follows by using the definition of V

(t −1). We now introduce
the quantity ˇ µ(x+t), given by
ˇ µ(x+t) =
J
x+t−1
/
x+t
. (5.8)
and the (modified) sum at risk
ˇ
R(t), which is given by
ˇ
R(t) =l
ad
(t) −(V

(t −1)(1+¯a(t)) −ì(t)) . (5.9)
By inserting Equations (5.8) and (5.9) into Equation (5.7), we obtain the
following formula:
V

(t) =(1+¯a(t)) V

(t −1) +(1−~(t))r(t) −ì(t) − ˇ µ(x+t)
ˇ
R(t). (5.10)
This recursive system differs from the system used for traditional with-profit
life insurance contracts in that the return ¯a(t) credited in year t to the policy
holder’s account is directly linked to the actual return from the underlying
investment portfolio. It is now essential to describe exactly how ì(t) and
¯a(t) should be chosen. We consider two different situations and explain how
financial mathematics may be applied when designing these contracts.
For example, a yearly guarantee o(t) can be specified by letting
¯a(t) =max(a(t). o(t));
5.3 The policy holder’s account 155
see also Equation (5.4). In this situation, o(t) is interpreted as the guaranteed
return between time t −1 and time t, and ì(t) is the payment for this guarantee.
In this situation, we can give simple explicit expressions for fair choices of
ì(t); we require that ì(t) is chosen at time t −1. If we alternatively consider
a terminal guarantee G(T), which may depend on the premiums paid up
to time T, we get a much more complicated product. Denote by G(t) the
terminal guarantee at time t, payable upon survival to T; the case G(t) =−
corresponds to the situation where there is no terminal guarantee. Nielsen and
Sandmann (1996) propose the following guarantee:
G(t) =
t
¸
;=0
r(;)(1−~(;))e
o

(T−t(;))
.
which means that the guarantee comprises the net premiums accumulated with
the inflation rate o

. In this situation, it is no longer obvious how ì(t) should
be chosen (or interpreted). The situation becomes more complicated if the
customer’s possibilities for changing investments or to stop paying premiums
are also included.
5.3.1 The financial market
We consider here the usual Black–Scholes model described in Chapter 4,
with two traded assets: a stock index S
1
and a savings account S
0
. The two
assets are given by
S
1
(t) =exp

o−
1
2
u
2

t +uW(t)

.
S
0
(t) =e
rt
.
where W = (W(t))
t∈|0.T ]
is a standard Brownian motion. We introduce the
u-algebra (t) =u]S
1
(u)u ≤t] which contains information about the devel-
opment of the stock up to and including time t. Let Q be a martingale measure
for S
1
¡S
0
, so that
S
1
(t) =exp

r −
1
2
u
2

t +uW
Q
(t)

.
where the parameter o has been replaced by r and the Brownian motion W
has been replaced by a Q-standard Brownian motion W
Q
. Finally, recall that,
for t ≤u,
E
Q
|e
−r(u−t)
S
1
(u)(t)] =S
1
(t). (5.11)
156 Integrated actuarial and financial valuation
5.3.2 Yearly guarantees and premiums
With the yearly guaranteed return o(t), ¯a(t) =max(a(t). o(t)), and hence the
forward recursion of the policy holder’s account becomes
V

(t) =V

(t −1)(1+a(t)) +(o(t) −a(t))
+
V

(t −1)
+(1−~(t))r(t) −ì(t) − ˇ µ(x+t)
ˇ
R(t). (5.12)
We propose to choose the payment ì(t) for the guarantee in year t so that
V

(t −1) =e
−r
E
Q
¸
V

(t) −r(t)(1−~(t)) + ˇ µ(x+t)
ˇ
R(t)

(t −1)
¸
.
(5.13)
where Q is a martingale measure. The interpretation of Equation (5.13) is as
follows. The value of the account at time t −1 corresponds to the value of the
account at time t, reduced by premiums paid at t and added risk premiums
ˇ µ(x +t)
ˇ
R(t). By inserting Equation (5.12) into Equation (5.13), we obtain
the following condition:
e
r
V

(t −1) =V

(t −1)

E
Q
| (1+a(t)) (t −1)]
+ E
Q
¸
(o(t) −a(t))
+

(t −1)
¸
−ì(t). (5.14)
where we have used that ì(t) is chosen at time t −1, i.e. that ì(t) is (t −1)-
measurable. Since Q is a martingale measure, and since
a(t) =|S
1
(t) −S
1
(t −1)]¡S
1
(t −1).
we see from Equation (5.11) that
E
Q
| (1+a(t)) (t −1)] =1+E
Q
| a(t) (t −1)] =1+e
r
−1 =e
r
.
From Equation (5.14), we now obtain the following expression for ì(t):
ì(t) =V

(t −1)E
Q
¸
(o(t) −a(t))
+

(t −1)
¸
. (5.15)
Explicit expression for the price of the guarantee
The price ì(t) of the guarantee can be determined explicitly by using the
Black–Scholes formula for the price of a European call option. We assume
that o(t) is chosen based on the information available at time t −1, that is
o(t) is (t −1)-measurable. Firstly, note that
(o(t) −a(t))
+
=

(1+o(t)) −
S
1
(t)
S
1
(t −1)

+
.
5.3 The policy holder’s account 157
Using the Black–Scholes formula (see Chapter 4), we then see that
E
Q
¸
S
0
(t −1)
S
0
(t)

(1+o(t)) −
S
1
(t)
S
1
(t −1)

+

S
1
(t −1)

=(1+o(t))e
−r
4(−z
2
(t)) −4(−z
1
(t)).
where 4 is the standard normal distribution function, and where
z
1
(t) =
−log(1+o(t)) +(r +
1
2
u
2
)
u
.
z
2
(t) =
−log(1+o(t)) +(r −
1
2
u
2
)
u
.
This shows that the fair price for the guarantee for the period (t −1. t] is
given by
ì(t) =V

(t −1) ((1+o(t))4(−z
2
(t)) −e
r
4(−z
1
(t))) . (5.16)
We note that this quantity only depends on the past development in the value of
the stock index via V

(t −1), which is the value of the policy holder’s account
at the beginning of the period |t −1. t]. In addition, the above calculations
show that the different periods can be considered separately when the price
of the guarantee is computed. In particular, this has the advantage that one
does not have to work with a model which describes the development of the
stock index until the time of maturity for the calculation of prices for the
guarantees.
Value of account payable upon death
We consider the special case
ˇ
R(t) =0, i.e. the situation where
l
ad
(t) =V

(t −1)(1+¯a(t)) −ì(t). (5.17)
This is the natural situation, where the value of the account is payable upon
death. Under this assumption, the recursion for the value of the account
simplifies as follows to:
V

(t) = V

(t −1)(1+a(t)) +(o(t) −a(t))
+
V

(t −1)
+(1−~(t))r(t) −ì(t). (5.18)
158 Integrated actuarial and financial valuation
It is not too difficult to see that Equation (5.18) is solved by
V

(t) =
t
¸
;=0
r(;)(1−~(;))
t
¸
s=;+1
(1+a(s))
+
t
¸
;=1

(o(;) −a(;))
+
V

(; −1) −ì(;)

t
¸
s=;+1
(1+a(s)). (5.19)
for t = 1. . . . . T. This expression shows how the value of the account can
be decomposed into two terms. The first term consists of net premiums
accumulated with the actual return on S
1
, and the second term is related to
the guarantees. One can show that Equation (5.15) implies that
T−1
¸
;=0
e
−r;
;
¡
x
(1−~(;))r(;)
=E
Q
¸
T
¸
;=1
;−1
¡
x 1
q
x+;−1
e
−r;
l
ad
(;) +
T
¡
x
e
−rT
V

(T)

. (5.20)
where V

(T) is given by Equation (5.19). This shows that the market value of
net premiums is equal to the market value of the benefits, so that the contract
is fair.
In more general models with stochastic interest rates, the market value
of the premiums appearing on the left hand side of Equation (5.20) would
involve the zero coupon bond prices P(0. ;) instead of the terms e
−r;
. In
the expression for the market value of the benefits (the right hand side of
Equation (5.20)), the factors e
−r;
and e
−rT
would have to be replaced by the
relevant stochastic discount factors, 1¡S
0
(;) and 1¡S
0
(T).
Account payable upon death under a pure unit-linked contract
We can simplify matters further by omitting the guarantee. Formally, this
corresponds to setting o(t) = −1, but it basically amounts to leaving out
the terms with ì(t) and (o(t) −a(t))
+
in the recursion for the value of the
account. This yields the following natural expression for the account:
V

(t) =
t
¸
;=0
r(;)(1−~(;))
t
¸
s=;+1
(1+a(s)) =
t
¸
;=0
r(;)(1−~(;))
S
1
(t)
S
1
(;)
.
(5.21)
i.e. the value of the account is simply the net premiums accumulated with
the actual returns from the index S
1
. This contract includes no guarantees on
the returns and the benefit upon death is the current value of the account.
Equation (5.21) indicates that this kind of contract is essentially a “bank
5.3 The policy holder’s account 159
product”: the value of the account is simply the premiums accumulated with
“interest,” and this amount is payable at the time of death or at maturity T if
the policy holder survives to T.
5.3.3 Terminal guarantees
We now turn to the situation where the return a(t) is credited to the pol-
icy holder’s account, but where the amount payable at time T is given
by max(V

(T). G(T)). For simplicity, we keep the assumption that the
value of the account is payable upon death, which ensures that
ˇ
R(t) = 0;
see Equation (5.17). The value of the account now satisfies the following
equation:
V

(t) =V

(t −1)(1+a(t)) +(1−~(t))r(t) −ì(t). (5.22)
As above, V

(t) can be written as follows:
V

(t) =
t
¸
;=0
r(;)(1−~(;))
t
¸
s=;+1
(1+a(s)) −
t
¸
;=1
ì(;)
t
¸
s=;+1
(1+a(s)).
and the question is howì(1). . . . . ì(T) should be chosen in order to ensure that
the contract is fair. More precisely, this amounts to finding ì(1). . . . . ì(T)
such that the market values of benefits and premiums balance. Under a
terminal guarantee, we obtain the following equation:
T−1
¸
;=0
e
−r;
;
¡
x
(1−~(;))r(;)
=E
Q
¸
T
¸
;=1
;−1
¡
x 1
q
x+;−1
e
−r;
l
ad
(;) +
T
¡
x
e
−rT
max(V

(T). G(T))

.
(5.23)
which is similar to Equation (5.20) for the case of yearly guarantees. Here,
the last term, max(V

(T). G(T)), can be quite complicated. One possibility
is to choose ì(t) =ì (constant) and determine
E
Q
¸
e
−rT
max(V

(T). G(T))
¸
=E
Q
¸
e
−rT
max

T
¸
;=0
(r(;)(1−~(;)) −ì)
T
¸
s=;+1
(1+a(s)). G(T)

.
where we use r(T) =0. Equation (5.23) can then be solved via simulation.
A similar problem is described in Bacinello and Ortu (1993a), who apply fix
point theorems in order to find fair parameters.
160 Integrated actuarial and financial valuation
5.3.4 Pure endowment paid by single premium
A unit-linked pure endowment paid by a single premium at time 0 is obtained
by letting r(1) = · · · = r(T −1) = 0 and l
ad
(1) = · · · = l
ad
(T) = 0. This
contract is fair if the market value of the benefits is equal to the market value
of the premiums, i.e. if
(1−~(0))r(0) =
T
¡
x
E
Q
¸
e
−rT
max(V

(T). G(T))
¸
.
where V

(T) is defined by the recursion, Equation (5.10). The terminal value
V

(T) depends on the development of the stock index, the net single premium
(1−~(0))r(0) and the price ì of the guarantee. Using the notation introduced
in Section 5.2, we can rewrite the amount payable at T as follows:
1(S
1
r(0). ì) =max(V

(T). G(T)). (5.24)
where we have chosen to underline the fact that the benefit depends on the pair
(r(0). ì). In the rest of this chapter we study unit-linked pure endowments
with benefits of the form given in Equation (5.24) without specifying 1
explicitly.
5.3.5 On the choice of model
In the previous sections, we have worked with the standard Black–Scholes
model. It is well known that this model does not give a perfect description of
the development of prices of financial assets. However, the model can still be
used to provide an important insight into how premiums, benefits and payment
profiles interact. For example, one can apply the model for comparisons of
various payment profiles under various contracts with different guarantees.
In addition, one can simulate developments of the index and compare the
distribution of the benefits for various choices of guarantees.
So what is wrong with the Black–Scholes model? One aspect is the assump-
tion of a constant interest rate on the savings account. For long term consid-
erations, it would certainly seem more natural to include a stochastic interest
rate model.
Another aspect is the choice of model for the returns on the underlying
stock index. According to the model, the daily changes in the logarithm of the
index should be independent and normally distributed with the same variance.
Empirical evidence shows that this is typically not the case for financial time
series. Firstly, financial time series typically show more heavy tails, which
means that there are more large changes in the real world than predicted by
the Black–Scholes model. This implies that the assumption of log-normally
5.4 Hedging integrated risks under diversification 161
distributed stock returns underestimates the risk of large losses in the index.
Secondly, financial time series typically show changes in the underlying
variance (the volatility), such that there are periods with large changes and
more quiet periods. This phenomenon, known as stochastic volatility, is also
not captured by the Black–Scholes model.
A reference to unit-linked life insurance contracts which investigates these
aspects more closely is Hardy (2003). For the modeling of extreme events in
finance and insurance in general, see Embrechts, Klüppelberg and Mikosch
(1997) and references therein.
5.4 Hedging integrated risks under diversification
A simple valuation formula for unit-linked life insurance contracts of the form
in Equation (5.1) can be derived under the assumption that the company can
invest in financial instruments (options) which pay 1(S
1
) at time T by using
a diversification argument corresponding to the one used in Section 3.2.6.
We repeat this argument here in a slightly modified version which also takes
mortality risk into account. Denote by r(0. 1) the price at time 0 of the option
which pays 1(S
1
) at time T. This contract essentially plays the role of the
zero coupon bond from the calculation of the market value of the guaranteed
payments on a traditional life insurance contract.
Assume that the insurance company purchases at time 0 exactly « options
for each policy holder, that is /
x
« options in total. This investment generates
the payoff /
x
«1(S
1
) at time T, so that the present value at time 0 of the
company’s net loss (expenses minus income) is given by
¯
L =Y(T)e

T
0
r(u) du
1(S
1
) −/
x
r(0)

/
x
«1(S
1
)e

T
0
r(u) du
−/
x
«r(0. 1)

. (5.25)
where
Y(t) =
/
x
¸
i=1
1
]T
i
>t]
is the actual number of survivors at time t ∈ |0. T]. The first two terms in
Equation (5.25) represent the present value at time 0 of the net payment to
the policy holders, i.e. the net loss associated with the contract which appears
without taking into account the investment side. The last term is exactly the
present value of the net loss (or minus the net gain) from investing in options.
Thus,
¯
L is the net payment to the policy holders reduced by investment gains.
162 Integrated actuarial and financial valuation
By rewriting the total net loss as follows:
¯
L =(Y(T) −/
x
«)e

T
0
r(u) du
1(S
1
) +/
x
(«r(0. 1) −r(0)) .
and using the independence between the financial market and the insured
lives, we see that
E|
¯
L] =(/
x T
¡
x
−/
x
«)E
¸
e

T
0
r(u)du
1(S
1
)
¸
+/
x
(«r(0. 1 ) −r(0)).
Thus, E|
¯
L] =0 if for example, the single premium r(0) is determined such
that
r(0) =
T
¡
x
r(0. 1) (5.26)
and « =
T
¡
x
. Here, we have calculated the expected value by using the true
survival probabilities. In fact, the criterion leads to the fair premium for the
pure endowment, which is equal to the option price r(0. 1) multiplied by the
survival probability
T
¡
x
.
In the literature, the premium calculation principle from Equation (5.26) is
often said to be a consequence of the insurance company being risk-neutral
with respect to mortality. This notion can, for example, be explained by
the fact that the fair premium can be derived by using the law of large
numbers and the usual limiting argument, which applies provided that the
lifetimes of the policy holders are independent. More precisely, the premium in
Equation (5.26) and the corresponding choice of « are uniquely characterized
by the property that the relative loss (1¡/
x
)
¯
L converges to zero when the size
/
x
of the portfolio increases (to infinity). Thus, the fair premium is actually
derived via an argument which neglects the mortality risk present in any
(finite) life insurance portfolio. Another explanation of the notion “risk-neutral
with respect to mortality” could be that the factor
T
¡
x
in Equation (5.26) is
the true expected value of 1
]T
1
>T]
, i.e. the true survival probability, whereas
the second factor, r(0. 1), is the price of the option 1(S
1
) at time 0, which is
typically not equal to the expected value under P of the discounted payment.
A discussion of this principle and these concepts can also be found in Aase
and Persson (1994).
The fair premium r(0) has the property that lim
/
x

(1¡/
x
)
¯
L = 0, if the
lifetimes are independent, such that the insurance company (in principle)
is able to eliminate the total risk in a portfolio of unit-linked life insurance
contract by buying standard options on the stock and by increasing the number
of contracts in the portfolio. Similarly, we have that if the premium r

(0) is
5.5 Hedging integrated risk in a one-period model 163
larger than that in Equation (5.26), then, for « =
T
¡
x
,
lim
/
x

1
/
x
¯
L =
T
¡
x
r(0. 1) −r

(0) -0.
and hence
¯
L →−. This leads to an infinitely large surplus, when the size
of the portfolio is increased, if the single premium is larger than the product
of the option price r(0. 1) and the probability
T
¡
x
of survival to T. Lower
premiums lead of course to an infinitely big loss as the size of the portfolio
is increased.
In the above analysis, we have obviously simplified matters greatly by
disregarding various facts of both theoretical and practical importance. One
example is the difference in time horizons between life insurance contracts
(typically more than fifteen years) and standard options (typically less than
one year), and one can certainly discuss whether it makes sense to work with
options with maturities of fifteen years. Another example is the choice of
valuation basis for the calculation of premiums of a unit-linked insurance
contract, where one might be interested in applying the same first order
valuation principle for the insurance risk as the one applied in the pricing of
traditional life insurance contracts. However, one can still ask questions like:
is it optimal to purchase /
xT
¡
x
options which each pay 1(S
1
) at time T? Does
there exist a better investment strategy for issuers of unit-linked insurance
contracts? Is it possible to find reasonable investment strategies based on
the underlying stock index? Is it possible to currently update the investment
strategy as new information from the financial markets and from the portfolio
of insured lives becomes available?
In order to be able to address these issues, one has to specify optimization
criteria. In the following, we look at several different criteria that can be
applied when determining optimal investment strategies.
5.5 Hedging integrated risk in a one-period model
In this section we consider a one-period model, where it is possible to invest
in a stock and a bond at two times only: time 0 and time 1. Denote by S
1
(t)
the value of the stock at time t ∈ ]0. 1], and let S
0
(t) be the value at t of one
unit invested in the savings account at time t. We assume that S
0
(t) =(1+r)
t
,
where r >0 is the annual interest from time 0 to time 1, which is furthermore
assumed to be known at time 0. In contrast, the value of the stock S
1
(1) at
time 1 is known at time 1 only.
164 Integrated actuarial and financial valuation
Consider now an insurance company facing a liability payable at time 1,
with present value H at time 0, and assume that the insurance company
is interested in reducing the risk associated with this liability as much as
possible. (Here we need to be careful and specify exactly how risk should be
measured; we come back to this later.) We start by describing the insurance
company’s possibilities in the financial market more precisely. This is again
formalized by introducing an investment strategy h. In the one-period model,
an investment strategy consists of a number of stocks h
1
, which are bought
at time 0 and are a part of the investment portfolio until time 1, where the
new price S
1
(1) is announced. In addition, the strategy involves the amount
h
0
(0), which is invested to or borrowed from the savings account at time
0. We work again with discounted prices defined by X(t) =S
1
(t)¡S
0
(t) and
X
0
(t) = S
0
(t)¡S
0
(t) = 1, respectively. The discounted value of the strategy
after the purchase of h
1
stocks at time 0 is given by
V(0. h) =h
1
X(0) +h
0
(0).
At time 1, the value of the stocks changes to h
1
X(1). If the insurance company
decides to change the deposit on the savings account at time t to h
0
(1), the
value at time 1 of the investment portfolio becomes
V(1. h) =h
1
X(1) +h
0
(1).
In order to ensure that the company is able to cover the liability H at time 1
by selling the stocks and by withdrawing all capital from the savings account,
it might be necessary to add additional capital to the savings account at
time 1. Similarly, if the value of the investments exceeds the liabilities, the
company may withdraw capital from the investment strategy before covering
the liability. More precisely, we require that h
0
(1) is chosen at time 1 so that
V(1. h) =H.
Föllmer and Sondermann (1986) introduced the cost process for the strat-
egy h in a model with trading in continuous time, and Föllmer and Schweizer
(1988) suggested a similar version in a discrete-time model. In the one-period
model considered above, the cost process is defined via C(0. h) = V(0. h)
and
C(1. h) =V(1. h) −h
1
(X(1) −X(0)) =H−h
1
AX(1). (5.27)
where we have used the usual notation AX(1) := X(1) −X(0), and where
the second equality is a consequence of the condition V(1. h) = H. More
precisely, the quantity C(t. h) represents the accumulated costs up to time t.
At time 0, the costs are equal to the initial investment V(0. h) corresponding
to the portfolio (h
0
(0). h
1
), and at time 1 the costs are computed as the value
5.5 Hedging integrated risk in a one-period model 165
V(1. h) = H of the new portfolio (h
0
(1). h
1
) reduced by investment gains
h
1
AX(1).
The costs given by Equation (5.27) are in fact closely related to
¯
L from
Equation (5.25), which we referred to as the company’s net loss. The quantity
¯
L was defined as the sum of the net payment to the policy holders (benefits
less premiums) and the net financial loss from trading in the financial market.
This means that Equation (5.27) essentially differs from the net loss
¯
L only
via the premiums paid at time 0.
Variance minimization
A simple measure of the risk associated with the liability H and the investment
strategy h is the variance of the accumulated costs at time 1. One idea
is therefore to choose an investment strategy, which solves the following
problem:
minimize
h
1
Var|C(1. h)]. (5.28)
One can refer to the solution as the variance-minimizing strategy. Prob-
lem (5.28) is so simple that it can be solved for a general liability without
imposing additional assumptions on the distribution of the discounted value
X(1) of the stock as time 1. Direct calculations show that
Var|C(1. h)] = Var|H−h
1
AX(1)]
= Var|H] −2h
1
Cov(H. AX(1)) +(h
1
)
2
Var|AX(1)] =: 1(h
1
).
where we have that h
1
is chosen at time 0 so that it is constant. Under the
condition Var|AX(1)] >0, we see that 1

(h
1
) >0, and hence 1 is minimized
for
¨
h
1
satisfying 1

(
¨
h
1
) =0; that is
¨
h
1
=
Cov(H. AX(1))
Var|AX(1)]
. (5.29)
By inserting this solution into the expression for the accumulated costs at
time 1, we see that the minimum obtainable variance is given by
Var|H−
¨
h
1
AX(1)] = Var|H] −
Cov(H. AX(1))
2
Var|AX(1)]
= Var|H]

1−Corr(H. AX(1))
2

. (5.30)
where
Corr(H. AX(1)) =
Cov(H. AX(1))

Var|H] Var|AX(1)]
166 Integrated actuarial and financial valuation
is the correlation coefficient between H and AX(1). The solution, Equa-
tion (5.29), and the minimum variance, Equation (5.30), are recognized as the
solution to the problem of minimizing the variance on a linear estimator.
Risk minimization
A similar problem is to minimize the expected value of the additional costs,
which occur between time 0 and 1, i.e.
minimize
(h
0
(0).h
1
)
E|(C(1. h) −C(0. h))
2
]. (5.31)
as a function of the number of stocks h
1
and the deposit h
0
(0) made on the
savings account at time 0. Since C(0. h) is constant and equal to V(0. h), we
see that this problem is solved by a strategy
¨
h with C(0.
¨
h) = E|C(1.
¨
h)],
so that the solution to Problem (5.31) is also variance minimizing. This
observation could lead to the (wrong) idea that this new problem does not
lead to new insight into the risk inherent in H. However, the advantage of
the new Problem (5.31) is that it also leads to an optimal initial investment
V(0.
¨
h), which is given by
V(0.
¨
h) =C(0.
¨
h) =E|H−
¨
h
1
AX(1)] =E|H] −
¨
h
1
E|AX(1)]. (5.32)
Föllmer and Schweizer (1988) refer to this quantity as the fair premium; the
strategy which solves Problem (5.31) is called the risk-minimizing strategy.
Under the condition E|AX(1)] = 0, which means that the discounted price
process X is a martingale, the fair premium is exactly equal to the expected
present value of the liability, i.e. V(0.
¨
h) =E|H]. However, if this condition
is not satisfied, the premium typically differs from the expected present value
of the liability.
Risk minimization for a unit-linked contract
Consider, for example, a portfolio of pure unit-linked pure endowments, i.e.
unit-linked contracts without any guarantees, where the policy holders simply
receive the value S
1
(1) of one unit of the stock index at time 1 provided that
they are still alive at this time. The present value at time 0 of the liability is
given by
H =Y(1)X(1).
where Y(1) =
¸
/
x
i=1
1
]T
i
>1]
is the actual number of survivors at time 1. At this
point, we are not assuming that the insured lifetimes are mutually independent.
However, if we assume that the lifetimes (T
1
. . . . . T
/
x
) are independent of the
5.5 Hedging integrated risk in a one-period model 167
stock X, we obtain, by using standard formulas for conditional covariances,
the following:
Cov(Y(1)X(1). AX(1)) = E|Cov(Y(1)X(1). AX(1) X(1))]
+Cov(E|Y(1)X(1) X(1)]. E|AX(1) X(1)])
= 0+ Cov(E|Y(1)] X(1). AX(1))
= E|Y(1)]Var|AX(1)].
By inserting this into the expression for the optimal investment strategy,
Equation (5.29), we obtain the optimal number of stocks:
¨
h
1
=E|Y(1)] =/
x 1
¡
x
=/
x+1
.
Thus, it is optimal to buy a number of stocks that corresponds to the expected
number of survivors. The optimal initial investment given by Equation (5.32)
is given by
V(0.
¨
h) = E|H] −
¨
h
1
E|AX(1)]
= E|Y(1)]E|X(1)] −E|Y(1)]E|X(1) −X(0)]
= E|Y(1)]X(0).
This result is similar to the fair premium suggested in Section 5.4, since the
price at time 0 of a contract which pays one unit of the stock index at time
1 is exactly S
1
(0) =X(0) (buy the stock at time 0 and sell it again at time 1).
The minimum obtainable variance (or the minimal risk) associated with
the risk-minimizing strategy can now be easily determined by exploiting the
independence between Y(1) and X(1). By using formulas for the calculation
of conditional variances, we obtain the following:
Var|H] = E|Var|Y(1)X(1)X(1)]] +Var|E|Y(1)X(1)X(1)]]
= E|X(1)
2
]Var|Y(1)] +(E|Y(1)])
2
Var|X(1)]. (5.33)
Now insert this into the expression for the minimum obtainable variance,
Equation (5.30), to see that
Var|C(1.
¨
h)] =E|X(1)
2
]Var|Y(1)]. (5.34)
168 Integrated actuarial and financial valuation
When is mortality risk diversifiable?
Let us study the behavior of the minimal risk as /
x
increases by considering
the standardized liability:
H =
1
/
x
Y(1)X(1) =
1
/
x
/
x
¸
i=1
1
]T
i
>1]
X(1). (5.35)
We say that mortality risk is diversifiable if the minimum risk for Equa-
tion (5.35) converges to zero as /
x
converges to infinity. For example, one
typically assumes that the policy holders’ lifetimes are independent of the
stock index. Is this sufficient to ensure that mortality risk is diversifiable? We
study this problem in more detail in the following.
Diversifiable mortality risk: independent lifetimes
If, in addition, we assume that the lifetimes T
1
. . . . . T
/
x
are mutually inde-
pendent and identically distributed, and that they are independent of the stock
index, we see that Y(1) ∼Bin(/
x
.
1
¡
x
), and hence
Var|Y(1)] =/
x 1
¡
x
(1−
1
¡
x
).
By replacing Y(1) by (1¡/
x
)Y(1) in the calculations leading to Equation (5.34),
we find that the minimum obtainable risk, Equation (5.30), for Equation (5.35)
is given by
E|X(1)
2
]
1
/
x
1
¡
x
(1−
1
¡
x
).
which indeed converges to zero as /
x
converges to infinity. In this situation,
mortality risk is diversifiable.
Non-diversifiable mortality risk
It is not difficult to realize that mortality risk is not diversifiable in the
general case. To see this, consider as a trivial example the case where all
lifetimes are identical, i.e. T
i
=T
1
, i =1. 2. . . . In this case, Equation (5.35)
is in fact identical to 1
]T
1
>1]
X(1), which does not depend on /
x
. A more
interesting example is obtained by assuming that the insured lifetimes are
independent given some underlying random variable 0, say, which is related to
the mortality intensity. For example, one could assume that the true mortality
intensity ¯µ(x +t) is unknown and is of the form 0µ(x +t), where µ(x +t)
is a known function and where 0 is a non-observable random variable with
some given distribution. One can specify this classical credibility model in
the following way.
5.5 Hedging integrated risk in a one-period model 169
• Given 0 = 6, T
1
. T
2
. . . . are independent and identically distributed with
survival probability
1
¡
(6)
x
=P
6
(T
1
>1) =exp

1
0

x+s
ds

.
• 0 and T
1
. T
2
. . . . are independent of the stock index X.
• 0 follows a non-degenerate distribution U.
This mortality model was used by Norberg (1989) for the analysis of a
portfolio of group life insurance contracts. Since the parameter 0 is not
known at time 0, one could choose to apply the survival probability E|
1
¡
(0)
x
]
when valuating life insurance contracts. However, by applying the law of
large numbers, we see that
1
/
x
Y(1) =
1
/
x
/
x
¸
i=1
1
]T
i
>1]
→P
0
(T
1
>1) =exp

1
0

x+s
ds

as /
x
goes to infinity. This limit is no longer constant and typically differs from
E|
1
¡
(0)
x
]. Thus, the problem is that by using the survival probability E|
1
¡
(0)
x
]
one systematically applies a wrong probability, whereas the true conditional
(and unknown) probability is given by
1
¡
(0)
x
. The financial consequences
as the size of the portfolio is increased are obvious and dramatic: if the
conditional survival probability exceeds the one used, the insurance company
systematically loses money, and if the conditional probability is smaller than
the one used, the company will systematically earn money.
By using the fact that Y(1), given 0 = 6, is binomially distributed with
parameters (/
x
.
1
¡
(6)
x
), and by using the usual formulas for calculation of
conditional variances, we see that
Var|Y(1)] = E|Var|Y(1)0]] +Var|E|Y(1)0]]
= /
x
E
¸
1
¡
(0)
x
(1−
1
¡
(0)
x
)
¸
+/
2
x
Var
¸
1
¡
(0)
x
¸
. (5.36)
In order to calculate the minimum obtainable risk, Equation (5.30), for
the standardized liability given in Equation (5.35), one has to determine
Var|(1¡/
x
)Y(1)]. Here, one sees that the minimum obtainable risk becomes
E|X(1)
2
]
1
/
2
x
Var|Y(1)] = E|X(1)
2
]

1
/
x
E
¸
1
¡
(0)
x
(1−
1
¡
(0)
x
)
¸
+Var
¸
1
¡
(0)
x
¸

.
which converges to
E|X(1)
2
] Var
¸
1
¡
(0)
x
¸
(5.37)
as /
x
goes to infinity. Thus, the last term in Equation (5.36) determines the
asymptotic behavior of the risk. We see that the company’s relative risk does
not converge to zero, even if the size of the portfolio increases. This is in
170 Integrated actuarial and financial valuation
line with the observations made above and is of course not surprising, since
we are in a situation where 0 is unknown. A similar observation can be
made from the limit given in Equation (5.37) for the minimal risk, which
exactly describes the uncertainty associated with the survival probability.
The risk associated with this underlying randomness is also called systematic
mortality risk, since it cannot be eliminated by increasing the size of the
portfolio. This is in contrast to the traditional mortality risk, the unsystematic
mortality risk, which vanishes when the portfolio is increased.
The example can be generalized further by assuming that the mortality
intensity is affected by some process 0 = (0(t))
t≥0
, such that the mortality
intensity at time t is a function of 0(t). The process 0 determines social and
economic factors that affect the mortality in the portfolio of insured lives. In
the above example, one only needs to change the definition of the conditional
survival probability to the following:
t
¡
(0)
x
=exp

t
0
µ
x+s
(0(s)) ds

.
Dahl (2004) analyzes the implications of allowing the mortality intensity to
depend on some underlying stochastic processes that are driven by diffusion
processes and determines market values for life insurance liabilities in this
setting. In Dahl and Møller (2006), this setup is applied for determining
risk-minimizing hedging strategies.
Non-diversifiable mortality risk and catastrophes
Further examples with non-diversifiable mortality risk can be constructed
by allowing for single events which may cause simultaneous deaths for any
fraction of the insured lives (catastrophes). A simple example is the situation
where a catastrophe occurs between time 0 and 1 with probability ¡
0
and
where the number of deaths caused by this catastrophe is binomially dis-
tributed with parameters (/
x
. ¡), where ¡ is uniformly distributed on (0, 1).
If this catastrophe is the only cause of death, it is not difficult to show that
the variance of (1¡/
x
)Y(1) does not go to zero as /
x
converges to infinity.
Modeling systematic mortality risk
As pointed out above, the economic consequences of the presence of the
systematic mortality risk are indeed dramatic. If, for example, we use survival
probabilities that are too low for the pricing of life annuities or pure endow-
ments, we systematically lose money. It is therefore necessary to study models
that take into consideration possible changes in the future mortality. How-
ever, since the future mortality is affected by many (unpredictable) factors,
5.5 Hedging integrated risk in a one-period model 171
one should use models where the future mortality is modeled via stochastic
processes.
Various approaches for the modeling of the systematic mortality risk have
been studied in the literature. One approach is the so-called Lee–Carter method
(see Lee (2000) and Lee and Carter (1992)), where the age-dependent yearly
death rates are affected by a time-dependent factor determined from a time
series model. Milevsky and Promislow (2001) aimed to model the mortality
intensitydirectlybyusinga so-calledmean-revertingGompertz model, whereas
Olivieri and Pitacco (2002) applied Bayesian methods to model the survival
probability. Dahl (2004) focused on an affine diffusion model for the mortality
intensity, and Biffis and Millossovich (2006) studied a general affine jump-
diffusion model. Further references of interest are Cairns, Blake and Dowd
(2006), who applied various methods known from interest rate theory to model
the mortality, Olivieri (2001) and Marocco and Pitacco (1998).
Trading with mortality risk
In the presence of systematic mortality risk, we cannot eliminate the risk asso-
ciated with the mortality by increasing the size of the portfolio. An alternative
way of controlling this risk is of course to introduce assets that are linked
to the development of the mortality and which can be traded dynamically.
One possibility could be to introduce assets that depend explicitly on the
behavior of the insured lives in the insurance portfolio. This would be similar
to what has been called dynamic reinsurance markets proposed by Delbaen
and Haezendonck (1989) and Sondermann (1991), and studied more recently
by Møller (2004). Applications in life insurance can also be found in Møller
(1998) and Steffensen (2001).
So-called mortality bonds and mortality swaps have been introduced; see,
for example, Blake, Cairns and Dowd (2006) for a detailed description of
their properties. A mortality bond is essentially a bond whose payments are
dependent on the development of the mortality in some specific group. One
example is to consider a certain cohort (for example, all individuals of a certain
age in a certain country) and let the payments be proportional to the current
number of survivors in this group. In this case, the payments to the owner
of the bond increase if the underlying individuals live longer than expected.
This means that the market value of the mortality bond increases if mortality
improves, and hence the mortality bond can provide a hedge against a general
decrease in the underlying mortality. A mortality swap provides a similar
construction. Here, an insurer can exchange (swap) the uncertain payments
within a portfolio of life annuitants with a fixed, certain payment stream by
entering a mortality swap directly with a reinsurer.
172 Integrated actuarial and financial valuation
5.6 The multi-period model revisited
5.6.1 Important concepts
In this section, we briefly review the multi-period model from Chapters 3
and 4 with dynamic investment strategies. This model is used in Section 5.7
for the analysis of the risk inherent in a portfolio of unit-linked life insurance
contracts. This section recalls the definitions of a self-financing strategy, the
value process, the cost process, arbitrage, martingale measure and attainability.
The reader can also consult Chapter 4 for more details. We fix a time horizon
T and let S
1
(t) be the value of the stock at time t ∈ ]0. 1. . . . . T]. The value at
time t of one unit deposited in the savings account at time 0 is assumed to be
given by S
0
(t) =(1+r)
t
, where r is the annual interest. The discounted prices
are defined by X(t) =S
1
(t)¡S
0
(t) and X
0
(t) =S
0
(t)¡S
0
(t) =1, respectively.
Here, we do not specify a specific model for S
1
, but one can, for example,
think of the binomial market, where the relative change in the value of
the stock can attain two different values. This example is studied further in
Section 5.6.3.
An investment strategy is a process h = (h
0
. h
1
) = (h
0
(t). h
1
(t))
t =0.1.....T
,
where h
1
(t) depends on the information available at time t −1 and h
0
(t)
can depend on the information available at t. We let (t) represent the
information available at t and assume that (s) ⊆ (t) for 0 ≤ s ≤ t ≤ T.
The quantity h
1
(t) denotes the number of stocks in the company’s investment
portfolio from time t −1 to time t, and h
0
(t) is the discounted deposit (negative
deposit corresponds to a loan) on the savings account at t. The value process
V(h) describes the value at any time t of the current portfolio (h
0
(t). h
1
(t));
it is given by
V(t. h) =(S
0
(t))
−1
(h
1
(t)S
1
(t) +h
0
(t)S
0
(t)) =h
1
(t)X(t) +h
0
(t). (5.38)
In Chapters 3 and 4 we introduced the concept of a self-financing strategy
in two slightly different versions. At first sight, one could perhaps form the
impression that there is a difference between these two definitions. Therefore
we show here that the two definitions are indeed equivalent. In Chapter 3, an
investment strategy was said to be self-financing if its cost process,
C(t. h) =V(t. h) −
t
¸
s=1
h
1
(s)AX(s). (5.39)
was constant (and equal to C(0. h) =V(0. h)). This amounts to saying that
V(t. h) =V(0. h) +
t
¸
s=1
h
1
(s)AX(s). (5.40)
5.6 The multi-period model revisited 173
The interpretation of Equation (5.40) is as follows. If h is self-financing, the
discounted value of the portfolio h(t) = (h
0
(t). h
1
(t)) at time t is exactly
equal to the initial value V(0. h) at time 0, to which are added the discounted
trading gains from the investment in the stocks. For the period |s −1. s], the
discounted gains are h
1
(s)AX(s), the number h
1
(s) of stocks purchased at
time s −1 multiplied by the change AX(s) =X(s)−X(s −1) in the discounted
value of the stock from time s −1 to time s. The accumulated costs C(t. h) for
the interval |0. t] are the value of the portfolio at t reduced by past investment
gains. Thus, the costs represent that part of the total value that has not been
generated by investment gains.
We now see that Equation (4.18) is a consequence of this definition. First,
note that Equation (5.40) corresponds to
AV(t +1. h) =V(t +1. h) −V(t. h) =h
1
(t +1)AX(t +1).
Using Equation (5.38), we see that
V(t +1. h) −V(t. h) = (h
1
(t +1) −h
1
(t))X(t) +h
1
(t +1)(X(t +1) −X(t))
+(h
0
(t +1) −h
0
(t)).
By comparing these two expressions for AV(t +1. h), we see that
h
1
(t +1)X(t) +h
0
(t +1) =h
1
(t)X(t) +h
0
(t).
which corresponds to Equation (4.18) if we multiply by S
0
(t) on both side of
the equality sign.
A self-financing strategy h is said to be an arbitrage strategy if V(0. h) = 0.
V(T. h) ≥ 0 P-a.s. (P-almost surely) and P(V(T. h) > 0) > 0. A probability
measure Q is said to be an equivalent martingale measure for X if Q agrees
with the probability measure P about whether any event has probability zero
or not and if X is a Q-martingale, i.e.
E
Q
|X(u)(t)] =X(t) (5.41)
for 0 ≤t ≤u ≤T. In Equation (5.41), the term E
Q
|X(u)(t)] is the expected
discounted value of the stock at time u calculated by using the measure Q
and by conditioning on the information available at time t. It was shown in
Section 3.7.7 that the existence of a martingale measure ensures the absence
of arbitrage possibilities.
An agreement between two parties gives rise to a discounted liability H at
time T; we refer to H as a contingent claim. For a given contingent claim, it is
important to check if there exists a self-financing strategy h with V(T. h) =H.
If this is the case, we say that H is attainable. This concept is essential in our
174 Integrated actuarial and financial valuation
subsequent discussion on the choice of investment strategies for integrated
financial and insurance risks. The requirement of attainability says that there
exists a strategy h =(h
0
. h
1
) such that
H =V(T. h) =V(0. h) +
T
¸
t=1
h
1
(t)AX(t). (5.42)
Here, V(0. h) is the replication price, i.e. the amount which has to be invested
at time 0. By following the strategy h, the investor can generate sufficient
capital at time T in order to pay the liability H. We can see how V(0. h) can
be determined by using the martingale measure Q: since X is a Q-martingale,
taking expected values (under Q) on both sides in Equation (5.42) shows that
E
Q
|H] =E
Q
|V(T. h)] =V(0. h);
see Section 3.7.7 for more details.
5.6.2 Unattainable claims
Attainability is a very special property. As mentioned in Chapter 4, models in
which all contingent claims are attainable are also called complete. Important
examples mentioned in Chapter 4 are the binomial market and the Black–
Scholes market. In these two models, any contingent claim is attainable.
However, here it is absolutely essential that a contingent claim is defined
as a quantity which depends on the risk from the financial market only. For
example, a European call option (S
1
(T) −K)
+
can be hedged in both the
binomial and the Black–Scholes markets. In order to be able to distinguish
such claims from more complicated claims, we also use the notion purely
financial contingent claim.
The situation changes drastically if we consider contingent claims that
also depend on other sources of uncertainty (risk), for example on the number
of surviving policy holders from an insurance portfolio. It seems intuitively
reasonable that one cannot hedge such liabilities perfectly by trading in the
financial markets only if, for example, the liabilities depend on the number
of survivors. We also refer to a claim that depends on both insurance risk
and financial risk as an integrated contingent claim. Thus, the extension of
the set of contingent claims typically leads to a so-called incomplete market,
i.e. a market which is no longer complete, since these integrated claims are
not in general attainable.
The rest of this chapter presents an introduction to methods for determining
optimal investment strategies in incomplete markets in the case where the
liability H is not attainable, i.e. in situations where it is not possible to find a
5.6 The multi-period model revisited 175
replicating strategy for the liability. In this case, one cannot use the principle
of no arbitrage alone in calculating unique prices or replicating investment
strategies for the liability. This means, for example, that we cannot determine
a unique price for a unit-linked insurance contract within the present model
if we leave out the assumption of diversifiable mortality risk. However, there
are several other principles (theories) which can be applied for valuation and
hedging in such situations. Here, we give an introduction to these principles
and illustrate how they work (and how they do not work) for a portfolio of
unit-linked life insurance contracts.
5.6.3 The binomial model
In this section we recall the binomial model, suggested by Cox, Ross and
Rubinstein (1979), which is complete such that all purely financial claims are
attainable. This model was also described in Chapter 4; see also Baxter and
Rennie (1996) and Pliska (1997). For a presentation with emphasis on the
mathematical aspects of this model, see Shiryaev et al. (1994).
In the binomial model, the dynamics of the stock is given by
S
1
(t) =(1+Z(t))S
1
(t −1). (5.43)
where Z(1). . . . . Z(T) is a sequence of i.i.d. random variables with Z(1) ∈
]o. l] and where 0 -P(Z(1) =l) -1. The quantity Z(t) is the relative change
in the value of the stock during the period (t −1. t], and we assume that this
quantity is not known before time t. As mentioned above, the savings account
is described by S
0
(t) =(1+r)
t
. A natural condition on the parameters o. l. r
is to require that −1 - o - r - l, which means that the return on the stock
exceeds the interest with positive probability and vice versa. The discounted
price process X(t) =S
1
(t)¡S
0
(t) is given by
X(t) =X(t −1)
1+Z(t)
1+r
.
Introduce now the filtration G=((t))
t∈]0.1.... .T]
for S
1
given by
(t) =u]S
1
(1). . . . . S
1
(t)]
where (0) is trivial. Again, (t) can be interpreted as the information which
corresponds to observing the price of the stock S
1
up to time t. We define
a new probability measure Q by Q(Z(1) =l) =(r −o)¡(l −o) =q and by
requiring that Z(1). . . . . Z(T) are i.i.d. under Q.
176 Integrated actuarial and financial valuation
The assumption o - r - l ensures that 0 - q - 1, so that Q is indeed a
probability measure equivalent to P. Under Q, we see that
E
Q
|(1+Z(t))(t −1)] =1+E
Q
|Z(t)] =1+ql +(1−q)o =1+r.
where we used the independence between Z(t) and (S
1
(1). . . . . S
1
(t −1)) in
the first equality. This shows that the discounted price process X is a
(G. Q)-martingale, since
E
Q
| X(t) (t −1)] =X(t −1)
1
1+r
E
Q
| (1+Z(t)) (t −1)] =X(t −1).
In Chapter 4 it was shown that the binomial model is complete. We can
formulate this in the following way.
Theorem 5.1 Let H be a (discounted) purely financial contingent claim
which is payable at time T and assume that the financial market is modeled
by the binomial model. Then there exists a process o(H), such that
H =H(0) +
T
¸
;=1
o(;. H)AX(;). (5.44)
where H(0) is constant and where o(;. H) depends on the information
available at time ; −1.
Equation (5.44) is also called a representation formula, since it allows for
H to be represented as the sum of a constant H(0) and some terms of the form
o(;. H)AX(;). We underline that this representation property is very special
for the binomial model. If, for example, we were to replace the quantities
Z(1). . . . . Z(T) in Equation (5.43) with random variables that could attain
three different values (a trinomial model), this result would no longer hold.
In the trinomial model it is typically not possible to find a representation of
this simple form for a contingent claim. Another important example in which
such representation results cannot be established is a contingent claim that
depends on an additional source of risk, for example the number of survivors
from a portfolio of insured lives.
Theorem 5.1 can be used to derive a self-financing strategy h = (h
0
. h
1
)
which replicates a purely financial contingent claimH. If we let h
1
(t) =o(t. H)
and define h
0
(t) via
V(t. h) =h
1
(t)X(t) +h
0
(t) =H(0) +
t
¸
;=1
o(;. H)AX(;).
5.7 Hedging integrated risks 177
we see that h is self-financing and that V(T. h) = H. Thus, we have con-
structed a self-financing strategy which replicates H. This self-financing
strategy requires an initial investment of H(0) at time 0, that is H(0) is
the replication price for H (see Section 5.6.1). In Chapter 4, o(H) was
derived for a European call option, i.e. for H:
H =
1(S
1
(T))
S
0
(T)
.
with 1(S
1
(T)) =(S
1
(T) −K)
+
. In the binomial model, o(t. H) can be deter-
mined in general by introducing the following process:
V

(t) =E
Q
|H (t)].
and then defining o(H) via
o(t. H) =
Cov
Q
(AV

(t). AX(t) (t −1))
Var
Q
|AX(t) (t −1)]
. (5.45)
5.7 Hedging integrated risks
We describe theories for hedging and pricing in incomplete markets and
demonstrate how these theories would apply for a portfolio of unit-linked life
insurance contracts. All examples build on the binomial market, so that any
purely financial contingent claim is attainable, i.e. it can be priced uniquely
and hedged perfectly via a self-financing strategy. However, integrated con-
tingent claims which depend on both insurance risk and financial risk are, in
general, not attainable, and here we need other methods.
Before describing these various methods, we recall the combined model,
which is used in the analysis of the integrated contingent claims. The model is
constructed by simply combining the binomial market and the usual actuarial
model for the lifetimes of the policy holders. This construction leads to an
incomplete market with many different martingale measures, which implies
that the choice of martingale measure is no longer a trivial problem. Thus, in
such models one cannot derive a unique arbitrage-free price from the assump-
tion of no arbitrage alone. There are typically many different prices that are
all consistent with the principle of no arbitrage, and thus one needs to intro-
duce subjective criteria when calculating prices. In the following, we describe
several different principles for pricing and hedging in incomplete markets:
super-hedging, risk minimization, mean-variance hedging, utility-indifference
178 Integrated actuarial and financial valuation
pricing and quantile hedging. We show how these different principles can be
used for valuation and partial hedging/control of the risk in a portfolio of
unit-linked life insurance contracts. Our presentation focuses on this appli-
cation, and we refer the reader to Cont and Tankov (2003) and Föllmer and
Schied (2002) for more general treatments of these principles.
5.7.1 The combined model
Consider again the insurance portfolio consisting of /
x
policy holders aged
x at time 0, and let Y(t) be the number of policy holders that are alive at t.
The present value at time 0 of the liabilities from a T-year unit-linked pure
endowment is given by
H =
Y(T)1(S
1
(T))
S
0
(T)
. (5.46)
Here, the financial market is described by the binomial market, and the stock
price process S
1
is defined via Equation (5.43), that is S
1
(t) = (1 +Z(t))S
1
(t −1). We assume throughout that 1 is a non-negative function. The lifetimes
for the /
x
policy holders are assumed to be independent, identically distributed
and independent of the stock index S
1
.
Information
The information that is available to the insurance company at time t is now
described. It is assumed that the company at time t has access to information
concerning the number of deaths each year up to time t and knowledge about
the value of the stock at times 0. 1. . . . . t. This can be formalized by introduc-
ing the filtrations G=((t))
t∈]0.1.... .T]
, defined by (t) =u]S
1
(1). . . . . S
1
(t)],
and H=((t))
t∈]0.1.... .T]
, defined by (t) =u]Y(1). . . . . Y(t)]. The first fil-
tration G describes the financial market and (t) is the information which
stems from observing the price of the stock S
1
up to time t. The other filtra-
tion H contains information about the number of survivors and (t) is the
full knowledge about the number of deaths from time 0 until time t. We are
interested in constructing investment strategies h = (h
0
. h
1
) that depend on
both types of information. In order to describe this more precisely, we con-
struct a third filtration F =((t))
t∈]0.1.... .T]
, given by (t) =(t) ∨(t) =
u((t) ∪(t)). This simply means that (t) contains the information (t)
from the financial market and the information (t) about the number of sur-
vivors. The information structure is illustrated in Figure 5.1. The insurance
company can now use the full information F when determining the investment
strategy h.
5.7 Hedging integrated risks 179
financial markets S
insurance company
F = ((t))t ∈{0,1,...,T }
insurance portfolio
G = ((t))t ∈{0,1,...,T}
H = ((t))t ∈{0,1,...,T}
h = (h
0
,h
1
)
Figure 5.1. Information diagram with the three filtrations F, G and H and the
investment strategy h.
Martingale measures
We consider a specific martingale measure Q, which is closely related to
the probability measure Q that was introduced for the binomial market in
Section 5.6.3. The probability Q considered here has the following properties:
(1) X is a Q-martingale, and under Q, Z(1). . . . . Z(T) are i.i.d. with
Q(Z(1) =l) =(r −o)¡(l −o);
(2) T
1
. . . . . T
/
x
are i.i.d. under Q with Q(T
1
>t) =P(T
1
>t) =
t
¡
x
;
(3) the lifetimes are independent of the financial market under Q.
The first property says that Q is similar to the probability measure introduced
in the section with the binomial market, and the second property says that
the probability measure Q corresponds to the true probability P, when we
consider the lifetimes only. Finally, the last property says that the lifetimes
are independent of the financial market under Q.
One problem with this combined model is that there are (infinitely!) many
martingale measures. We can see this by looking more carefully at the prop-
erties of the martingale measure Q defined above. In fact, we can construct
probability measures Q
n
which also satisfy the first condition, so that they are
indeed martingale measures, where the financial market is independent of the
lifetimes (the third condition), but where the survival probability in the second
condition is changed to some value 0 -
t
¡
(n)
x
-1. The details of this construc-
tion are given in Section A.1. of the Appendix. Thus, if one aims to calculate
the price of the unit-linked life insurance contract as the expected present
value under a martingale measure, we obtain for the martingale measure Q
n
180 Integrated actuarial and financial valuation
the following:
E
Q
n
|H] =E
Q
n
¸
Y(T)1(S
1
(T))
S
0
(T)
¸
=E
Q
n
¸
1(S
1
(T))
S
0
(T)
¸
E
Q
n
|Y(T)]
=E
Q
¸
1(S
1
(T))
S
0
(T)
¸
/
x T
¡
(n)
x
.
By choosing different equivalent martingale measures, we can now get any
survival probability
T
¡
(n)
x
between zero and unity. This shows that any price
in the open interval,

0. /
x
E
Q
¸
1(S
1
(T))
S
0
(T)
¸
.
is in accordance with the assumption of no arbitrage. We can formulate this
result alternatively in the following way: arbitrage-free pricing cannot tell
us which survival probability we should use when we can only trade in the
stock and the savings account. The survival probability can, for example,
be chosen to be equal to the true survival probability
T
¡
x
, close to (but not
equal to) zero, close to (but not equal to) unity, or some other value. This is
not very surprising, since we have not introduced the possibility of trading
with mortality risk in the financial market. The situation changes when assets
that depend on the number of survivors are introduced. If these assets are
defined in the right way, it might be possible to reestablish uniqueness of the
martingale measure, and arbitrage-free pricing leads again to unique prices
also for integrated contingent claims.
We end this section by recalling the result from Theorem 5.1, which shows
that the unique arbitrage-free price for the purely financial contingent claim
that pays 1(S
1
(T)) at time T is given by
r(t. 1 ) :=E
Q
¸
1(S
1
(T))
S
0
(T)

(t)
¸
=E
Q
¸
1(S
1
(T))
S
0
(T)
¸
+
t
¸
;=1
o(;. 1)AX(;).
(5.47)
where o(1) determines the hedge for 1(S
1
(T)). Thus, one can hedge the
option 1(S
1
(T)) by using a self-financing strategy which requires an initial
investment of r(0. 1) =E
Q
|1(S
1
(T))¡S
0
(T)] at time 0 and an investment in
o(t. 1 ) stocks during the period (t −1. t]. Note that we can apply any equiva-
lent martingale measure in Equation (5.47) for the calculation of r(t. 1 ). Since
1(S
1
(T)) is attainable, we would obtain the same result in Equation (5.47) if
we replaced Q by another martingale measure Q
n
.
5.7 Hedging integrated risks 181
5.7.2 Super-hedging
A very natural suggestion for the choice of investment strategy and premium
calculation principle is the idea of super-hedging suggested by El Karoui and
Quenez (1995). For a contingent claim, which cannot be hedged perfectly by
a self-financing strategy h, such that one cannot ensure that V(T. h) = H in
all scenarios, one could look for strategies where
V(T. h) ≥H (5.48)
in any scenario. Thus, if the seller of the contract uses a strategy satisfying
this condition, and if they can charge the amount necessary as the initial
investment, then there is no risk of a loss associated with the liability for the
seller. By following the self-financing strategy, the seller generates the capital
V(T. h), which by Equation (5.48) exceeds the liability H. We show below
that this very natural idea unfortunately leads to unreasonably high prices for
unit-linked life insurance contracts.
Definition 5.2 (Super-hedging) Find the cheapest self-financing strategy h,
such that V(T. h) ≥ H P-a.s. The minimum initial investment V(0), which
allows for a self-financing strategy satisfying Equation (5.48), is called the
super-hedging price (or super-replication price), and the associated strategy
is called the super-hedging strategy.
In particular, we are interested in the valuation and hedging of unattainable
claims, for example as in the case with the unit-linked life insurance contracts.
In such situations there are typically many martingale measures, and we let
Q be the set of all martingale measures. It can be shown that a super-hedging
strategy for H exists provided that
sup
Q

∈Q
E
Q

|H] -. (5.49)
i.e. the expected value of H under all martingale measures is bounded. The
solution to the super-hedging problem involves some rather complicated math-
ematics, see El Karoui and Quenez (1995), and we formulate the result without
going into details here. First, it is necessary to study the following process:
V(t) =ess.sup
Q

∈Q
E
Q

|H(t)]. (5.50)
This involves the calculation of the expected value of H given the information
available at time t under all martingale measures from the set Q, and V(t)
is then essentially defined as the supremum of all these different expected
182 Integrated actuarial and financial valuation
values. The next step is to derive a decomposition for the process V of the
form
V(t) =V(0) +
t
¸
;=1
h
1
(;)AX(;) −C(t). (5.51)
where C is adapted and non-decreasing and where h
1
is predictable, i.e. h
1
(;)
is (; −1)-measurable for ; = 1. . . . . T. One of the main results from the
theory on super-hedging is given in the following.
Theorem 5.3 Provided that Equation (5.49) is satisfied, the super-hedging
price of H is given by V(0) = sup
Q
E
Q

|H] and the super-hedging strategy
is determined by h
1
(t) =h
1
(t).
5.7.3 Super-hedging for unit-linked contracts
This section illustrates how super-hedging strategies can be derived for unit-
linked contracts of the form given in Equation (5.46) within the market
introduced in Section 5.7.1 by using the results from Section 5.7.2. The first
step is to check the condition given by Equation (5.49). By noting that the
number of survivors at T must be smaller than the number /
x
of policy holders
at time 0 (no additional contracts are signed), we see that
H =
Y(T)1(S
1
(T))
S
0
(T)
≤/
x
1(S
1
(T))
S
0
(T)
=H

.
This simple inequality is quite useful, since H

is a purely financial con-
tingent claim, which in the binomial market is attainable, and hence it can
be priced uniquely by the principle of no arbitrage. Moreover, we see from
Equation (5.47) of r(0. 1 ) that the expected value of H

is the same under
all equivalent martingale measures and equal to /
x
E
Q
|1(S
1
(T))¡S
0
(T)]. Thus,
we have shown that
sup
Q

∈Q
E
Q

|H] ≤E
Q
|H

] =/
x
r(0. 1 ).
which is finite provided that r(0. 1 ) is finite (and this is the case here). One
might say that this looks like a rather academic discussion. However, this
property is actually quite essential. For example, in the situation where H =
N(T)1(S
1
(T))¡S
0
(T) and where N is a Poisson process (which is independent
of S
1
), the condition would no longer be satisfied, and this modified contract
cannot be super-hedged. No matter how big the initial capital is and no matter
how this capital is invested, there is a (positive) probability of having less
capital than needed in order to pay H.
5.7 Hedging integrated risks 183
By the above argument, we see that Y(t) ≥Y(T), so that for any martingale
measure Q
n
:
E
Q
n
¸
Y(T)1(S
1
(T))
S
0
(T)

(t)
¸
≤Y(t)E
Q
n
¸
1(S
1
(T))
S
0
(T)

(t)
¸
=Y(t)r(t. 1 ).
where we have used the fact that Y(t) is known at time t, such that it can
be taken outside of the expectation. If we consider a sequence of martingale
measures (Q
n
)
n∈N
, where the survival probabilities converge to unity as n
goes to infinity, we see that we can get arbitrarily close to Y(t)r(t. 1 ). This
shows that
V(t) =Y(t)r(t. 1 ).
and it remains to find the representation given by Equation (5.51). By noting
that
AV(t) =Y(t −1)(r(t. 1 ) −r(t −1. 1 )) +r(t. 1 )(Y(t) −Y(t −1)).
and by using Equation (5.47), we see after some rearrangement that
V(t) =/
x
r(0. 1 ) +
t
¸
;=1
Y(; −1)o(;. 1 )AX(;) −
t
¸
;=1
r(;. 1)(−AY(;)). (5.52)
Here, we see that Y(; −1)o(;. 1 ) is (; −1)-measurable. The process Y is
decreasing since it counts the current number of survivors. This implies that
−AY(;) ≥0, such that the process
t
¸
;=1
r(;. 1 )(−AY(;))
is indeed non-decreasing. Thus, we can now use Theorem 5.3 to obtain the
following Proposition.
Proposition 5.4 The super-hedging price for the unit-linked contract is given
by /
x
r(0. 1 ). The cheapest self-financing super-hedging strategy for the unit-
linked contract is given by
h
1
(t) =Y(t −1) o(t. 1 ).
h
0
(t) =/
x
r(0. 1 ) +
t
¸
;=1
h
1
(;) AX(;) −h
1
(t)X(t).
where o(1 ) is determined from Equation (5.47).
184 Integrated actuarial and financial valuation
Let us comment on this result. We see that the price, which is necessary
in order to super-hedge the liability associated with the portfolio of unit-
linked life insurance contracts, is given by the initial number /
x
of policy
holders at time 0 multiplied by the price r(0. 1 ) of the purely financial
contingent claim 1(S
1
(T)). The investment strategy consists in holding from
time t −1 to time t a number of stocks which is calculated as the current
number Y(t −1) of survivors at time t −1 multiplied by o(t. 1 ). The latter is
exactly the number of stocks needed in order to hedge the claim 1(S
1
(T)).
A natural question is now: where did the survival probability go in all these
calculations? One answer is that the survival probability has been set to unity.
This result illustrates that the natural idea of super-hedging is not the right
tool for the handling of the integrated risk in a unit-linked insurance contract.
The requirement that the value of the portfolio at time T has to exceed H in
any scenario is simply too strong for our applications, since it implies that we
must have sufficient capital in order to cover even the case where all policy
holders survive to time T. We have to look for alternative criteria.
5.7.4 Risk minimization
Section 5.5 studied risk minimization in a one-period model. Here, we give
a version of this criterion in a multi-period model. For t ∈ ]0. 1. . . . . T −1],
we define the quantity
r(t. h) =E
Q
¸
(C(t +1. h) −C(t. h))
2

(t)
¸
. (5.53)
the conditional expected value at time t, under the martingale measure Q of
the square of the additional costs C(t +1. h)−C(t. h) incurred during the next
time interval. In contrast to Section 5.7.3, we here require that V(T. h) =H.
Since this is not in general possible under a self-financing strategy, we allow
for the possibility of adding or withdrawing capital at any time. However,
this additional capital is not being priced here.
Definition 5.5 (Risk minimization) Minimize r(t. h) for all t over h with
V(T. h) =H. The solution is called the risk-minimizing strategy.
Minimization of r(t. h) can be viewed as a generalization of Problem (5.31)
considered in the one-period model. However, Equation (5.53) is formulated in
terms of a specific martingale measure Q, a fact which we address below. The
idea of minimizing r(t. h) is in some sense analogous to the calculations for
the price in a complete market as described in Chapter 4, in that both problems
are solved via a backwards recursion. Here, we start by looking at r(T −1. h).
5.7 Hedging integrated risks 185
At time t, r(t. h) is minimized as a function of h
1
(t +1) and h
0
(t) for given
(h
1
(t +2). . . . . h
1
(T)) and (h
0
(t +1). . . . . h
0
(T)). This minimization can be
performed as in Section 5.5; for more details, see Föllmer and Schweizer
(1988). The approach of risk minimization was proposed by Föllmer and
Sondermann (1986), who applied a continuous-time setting. Surveys and
further references on risk minimization and other quadratic hedging criteria
can be found in Pham (2000) and Schweizer (2001a).
The risk-minimizing strategy can also be found by first considering the
process V

defined by
V

(t) =E
Q
|H (t)]. (5.54)
It can be shown that this process has a unique decomposition of the following
form:
V

(t) =V

(0) +
t
¸
;=1
h
1
(;. H)AX(;) +L(t. H). (5.55)
where h
1
(;. H) depends on the information available at time ; −1, and where
L(H) is a Q-martingale, which furthermore has the very special property that
the product of X and L(H) is also a Q-martingale, i.e.
E
Q
| X(u)L(u. H) (t)] =X(t)L(t. H)
for all t ≤ u. The decomposition, Equation (5.55), is known as the Kunita–
Watanabe decomposition; see Föllmer and Schied (2002) for more details. The
decomposition can typically be found by first determining Equation (5.54) and
then examining the increment AV

(t) = V

(t) −V

(t −1). In Section 5.7.5
we show how this works for a portfolio of unit-linked contracts.
It follows from Föllmer and Schweizer (1988) that the risk-minimizing
strategy
¨
h =(
¨
h
0
.
¨
h
1
) minimizing r (t. h) for all t is given by
¨
h
1
(t) =h
1
(t. H). (5.56)
¨
h
0
(t) =V

(t) −
¨
h
1
(t)X(t). (5.57)
Thus, the risk-minimizing strategy is determined directly from the decompo-
sition given in Equation (5.55). The proof is recalled in Section A.2. of the
Appendix. This strategy has the property that its value at time t is given by
V(t.
¨
h) =
¨
h
1
(t)X(t) +
¨
h
0
(t) =V

(t).
i.e. the value process coincides with the martingale, Equation (5.54). In par-
ticular, we see that the investment made at time 0 is V

(0) =E
Q
|H], which
can be interpreted as a fair price for H.
186 Integrated actuarial and financial valuation
Using the definition given in Equation (5.39) of the cost process, we see
that the cost process associated with the risk-minimizing strategy is given by
C(t.
¨
h) =V

(t) −
t
¸
;=1
¨
h
1
(;)AX(;) =V

(0) +L(t. H) (5.58)
and the minimum obtainable risk is given by
r (t.
¨
h) =E
Q
¸
(AL(t +1. H))
2

(t)
¸
.
Thus, in order to apply the criterion of risk minimization, it is only necessary
to derive the decomposition given in Equation (5.55), since this determines
the optimal strategy and its associated risk. Since L(H) is a martingale, we see
from Equation (5.58) that the cost process associated with the risk-minimizing
strategy is also a martingale. As a consequence, the risk-minimizing strategy
is also said to be mean-self-financing (self-financing in the mean). A self-
financing strategy is also mean-self-financing since the cost process of a self-
financing strategy is constant and hence, in particular, a martingale. However,
a mean-self-financing strategy is not in general self-financing.
Related criteria
The expected value in Equation (5.53) is calculated by applying the equivalent
martingale measure Q and not by using the original probability measure P,
although it would of course be much more natural to work with the probability
measure P. Here one has to be a bit careful, since this can lead to a problem
that has no solution; for example, this would be the case if Q is replaced by
P in Equation (5.59) below. However, one can show that the solution
¨
h also
minimizes
R(t. h) =E
Q
¸
(C(T. h) −C(t. h))
2

(t)
¸
(5.59)
for all t, over all strategies h with V(T. h) = H. With this formulation, the
quantity C(T. h) −C(t. h) can be interpreted as the future costs associated
with the strategy, and hence the criterion essentially amounts to minimizing
at any time the conditional variance on the total future costs.
An alternative problem is the following optimization problem:
minimize
h
1
(t).h
0
(t).t=0.1.... .T
E
Q
¸
T
¸
t=1
(C(t +1. h) −C(t. h))
2

. (5.60)
It can be shown that the solution h =(h
1
. h
0
) to Equation (5.60) only deviates
from
¨
h =(
¨
h
0
.
¨
h
1
) given by Equations (5.56) and (5.57) via the choice of h
0
and
¨
h
0
, i.e. h
1
=
¨
h
1
.
5.7 Hedging integrated risks 187
Alternatively, one could be interested in the strategy that minimizes, for
any t,
E
¸
(C(t +1. h) −C(t. h))
2

(t)
¸
. (5.61)
where the expected value under Q in Equation (5.53) has been replaced by the
true probability measure P. The solution to Equation (5.61) can be determined
via the following recursion formula taken from Föllmer and Schweizer (1988):
h
1
(t) =
Cov

H−
¸
T
;=t+1
h
1
(;)AX(;). AX(t)

(t −1)

Var|AX(t) (t −1)]
.
h
0
(t) = E
¸
H−
T
¸
;=t+1
h
1
(;)AX(;)

(t)

−h
1
(t)X(t).
As mentioned above, the criterion which consists of replacing Q by P in
Equation (5.59) does not lead to a well defined problem; see Schweizer
(1988, 2001a).
5.7.5 Risk minimization for unit-linked contracts
The main step in determining the risk-minimizing strategy for the unit-linked
contract is the derivation of the decomposition, Equation (5.55), within the
market introduced in Section 5.7.1. In order to be able to give a better
interpretation of our results, we introduce a process M defined by
M(t) =E
Q
| Y(T) (t)] =Y(t)
T−t
¡
x+t
. (5.62)
which keeps track of the conditional expected number of survivors at T given
the current number of survivors. The factor
T−t
¡
x+t
is the probability of
surviving to T given survival to t and Y(t) is the actual number of survivors
at t. Thus, the process fluctuates over time with the terminal condition M(T) =
Y(T), i.e. the terminal value of M coincides with the number of survivors at
T. One can express the decomposition given in Equation (5.54) for
V

(t) =E
Q
¸
Y(T)
1(S
1
(T))
S
0
(T)

(t)
¸
in terms of r(1 ) and M. More precisely, one can show that the decomposition
needed is given by
V

(t) =V

(0) +
t
¸
;=1
M(; −1)o(;. 1 )AX(;) +
t
¸
;=1
r(;. 1 )AM(;). (5.63)
188 Integrated actuarial and financial valuation
This follows by first using the independence between Y and (S
0
. S
1
) to obtain
V

(t) =M(t) r(t. 1 ).
which shows that
AV

(t) =M(t −1)(r(t. 1 ) −r(t. 1 −1)) +r(t. 1 )(M(t) −M(t −1)).
The decomposition in Equation (5.63) is now obtained by using Equa-
tion (5.47) for r(1 ); see Section A.3 of the Appendix for more details.
This leads to the following result, which can also be found in Møller (2001a).
Proposition 5.6 The fair price for the unit-linked contract is /
x T
¡
x
r(0. 1 ).
The risk-minimizing strategy is as follows:
h
1
(t) =Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1 ).
h
0
(t) =Y(t)
T−t
¡
x+t
r(t. 1 ) −Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1 )X(t).
where o(1 ) is determined by Equation (5.47).
We note the following. The optimal number of stocks h
1
(t) for the period
(t −1. t] has a very natural form: it is the product of the hedge o(t. 1 ) for
the option 1(S
1
(T)) and the conditional expected number of survivors,
Y(t −1)
T−(t−1)
¡
x+(t−1)
.
calculated at time t −1. In addition, we see that the deposit on the savings
account is currently being adjusted so that the value of the portfolio at t is
equal to V

(t). This risk-minimizing strategy is not self-financing, and this can
be seen by considering the cost process, which, according to Equation (5.58),
is given by
C(t. h) =V

(0) +L(t. H) =V

(0) +
t
¸
;=1
r(;. 1)AM(;).
Thus, the cost process is only constant if the process M defined by Equa-
tion (5.62) is constant or if r(t. 1 ) =0, which is typically not the case. The
quantity AM(t) is the difference between the expected number of survivors
calculated at time t and t −1, respectively. If the actual number of deaths
during the period (t −1. t] is smaller than the expected number of deaths, the
expected number of survivors increases, and hence AM(t) >0. This leads to
a loss for the insurance company, which must pay the amount 1(S
1
(T)) to
each of the survivors at T. The situation AM(t) -0 arises when the expected
number of survivors decreases, so that the company may reduce their reserves
5.7 Hedging integrated risks 189
for these contracts. If we write the survival probability
T−(t−1)
¡
x+(t−1)
in the
form
1
¡
x+(t−1) T−t
¡
x+t
, the loss for the period (t −1. t] can be rewritten as
follows:
AL(t) =r(t. 1 )
T−t
¡
x+t

Y(t) −Y(t −1)
1
¡
x+(t−1)

.
which has the following interpretation: that the insurance company’s loss is
proportional to r(t. 1 )
T−t
¡
x+t
, which is exactly the price at t of the option
1(S
1
(T)) multiplied by the survival probability
T−t
¡
x+t
. This quantity is the
market value at time t obtained by using the martingale measure Q for each
policy holder alive at t. The second factor (Y(t) −Y(t −1)
1
¡
x+(t−1)
) is the
difference between the actual number of survivors at t and the expected
number of survivors at t, calculated at t −1. It only remains to be shown
that Equation (5.63) is the decomposition we need, and this is verified in
Section A.3 of the Appendix. It was first suggested that risk minimization is
applied to unit-linked life insurance contracts by Møller (1998). An extension
that allowed for the handling of insurance payment streams was proposed by
Møller (2001c).
We end our treatment of risk minimization by assessing the risk which
cannot be hedged away, i.e. the minimum obtainable risk. For example, this
risk can be quantified by determining the variance under Q of the total costs
C(T. h) associated with the risk-minimizing strategy. Here, one can exploit
the fact that the increments of the martingales M and r(1) are uncorrelated,
and that the change of measure from P to Q does not affect the distribution
of the lifetimes to show that
Var
Q
|C(T. h)] =
T
¸
t=1
E
Q
|(r(t. 1))
2
AM(t)
2
]
=
T
¸
t=1
E
Q
|(r(t. 1))
2
]E|AM(t)
2
]. (5.64)
The term involving AM(t) can be expressed via the following survival prob-
abilities:
E|AM(t)
2
] = E|Var|AM(t) (t −1)]] +Var|E|AM(t) (t −1)]]
= E|Var|
T−t
¡
x+t
Y(t) (t −1)]]
= /
x T
¡
x T−t
¡
x+t
(1−
1
¡
x+(t−1)
). (5.65)
where the second equality follows by noting that M is a martingale and the
last equality follows by noting that Y(t) (t −1) ∼bin(Y(t −1).
1
¡
x+(t−1)
).
190 Integrated actuarial and financial valuation
The minimal variance given in Equation (5.64) should be compared with the
variance of H, which is given by
Var
Q
|Y(T)1(S
1
(T))¡S
0
(T)] = E
Q
|Var
Q
|Y(T)r(T. 1 )(T)]]
+Var
Q
|E
Q
|Y(T)r(T. 1 )(T)]]
= E
Q
|(r(T. 1 ))
2
]/
x T
¡
x
(1−
T
¡
x
)
+Var
Q
|r(T. 1 )](/
x T
¡
x
)
2
. (5.66)
This quantity is the variance of the company’s total loss in the situation with
no investment in the stock.
If we compare the risk-minimizing strategy and the super-hedging strategy,
we see that the risk-minimizing strategy has a more natural form in that it
involves the survival probability. However, one should be a bit careful here,
since we have actually made a subjective choice by choosing the martingale
measure Q in the calculations. Another martingale measure with an alternative
survival probability would lead to different strategies. In addition, one can
criticize the fact that the criterion of risk minimization is a quadratic criterion
which uses the square of the costs, so that losing 100 euro and earning
100 euro is equally bad! Therefore, it might be more desirable with a criterion
which only punished losses. Finally we mention that the criterion of risk
minimization leads to strategies that are typically not self-financing, i.e. it is
typically necessary currently to add and withdraw capital.
5.7.6 Mean-variance hedging
An alternative criterion is the principle of mean-variance hedging. This crite-
rion does not use a specific martingale measure, and it only uses self-financing
strategies. As with the criterion of risk minimization, mean-variance hedging
is a symmetric criterion that punishes gains and losses equally.
Definition 5.7 (Mean-variance hedging) Minimize E
¸
(H−V(T. h))
2
¸
over
all self-financing strategies h.
The principle consists in finding the self-financing strategy that comes as
close as possible to the liability H in the L
2
(P)-sense. Recall that the terminal
value at time T of the portfolio for a self-financing strategy is given by
V(T. h) =V(0. h) +
T
¸
;=1
h
1
(;)AX(;).
5.7 Hedging integrated risks 191
Thus, the principle of mean-variance hedging can be interpreted as the prob-
lem of projecting H on the linear subspace, which consists of all possible
values of a self-financing strategy. This can be formulated more precisely by
introducing the set of all these possible terminal values:
=]V(T. h) h self-financing] . (5.67)
The solution to the mean-variance problem is exactly the projection of
H on . Mathematically, one can exploit the fact that the problem can be
formulated within a Hilbert space (L
2
(P)), and, from the so-called projection
theorem, we know that H can be uniquely written as follows:
H =c(H) +
T
¸
;=1
h
1
(;. H)AX(;) +N(H) =o(H) +N(H). (5.68)
where o(H) ∈ and where N(H) is orthogonal to , in the sense that
E|N(H)o] =0 for all o ∈ . This is illustrated in Figure 5.2. The correspond-
ing initial capital c(H) is called the approximation price for H. Together with
h
1
(H), it determines the mean-variance strategy, i.e. the self-financing strat-
egy that minimizes the distance to H. The quantity N(H) is interpreted as the
part of the liability H which cannot be hedged. In Sections 5.7.8 and 5.7.9
we examine principles where the non-hedgeable part N(H) of the liability
appears in the premium in the form of a safety loading. There is a substantial
literature on this topic that involves quite complicated mathematical results;
key references are Rheinländer and Schweizer (1997) and Schweizer (2001a).
Here, we skip the technical details and formulate the main result directly
under some simplifying assumptions which are satisfied in our example of
the binomial market. However, even in this situation, we need to introduce
L
2
(P ) H
N(H )
a(H)

Figure 5.2. Mean-variance hedging. The approximation price and the optimal
investment strategy are determined as a projection.
192 Integrated actuarial and financial valuation
some additional notation in order to be able to give the solution. We define
AA(t) =E|AX(t)(t −1)]
for t =1. . . . . T, which is the expected change in X in year t computed under
the true probability measure P, and
¯
\(t) =
AA(t)
E|(AX(t))
2
(t −1)]
.
We can now define a measure via
d
¯
P
dP
=
T
¸
t=1
1−
¯
\(t)AX(t)
1−
¯
\(t)AA(t)
=
¯
Z(T). (5.69)
and use the notation
¯
E instead of E
¯
P
. In general, the density given by Equa-
tion (5.69) can take negative values, which implies that the measure
¯
P can
be a signed measure. Note that if X is a martingale under P, AA(t) =0 such
that A(t) = A(0) and
¯
\(t) = 0. In particular, this implies that P =
¯
P in this
case.
We assume here that the quantities
¯
\(t)AA(t) are deterministic, an assump-
tion which is verified for the binomial market in Section A.4 of the Appendix.
The calculations there also show that
¯
Z(T) is strictly positive and that
¯
P in
this case is in fact identical to the probability measure Q introduced in Sec-
tion 5.7.1.
Schweizer (1995, Proposition 4.3) shows that the mean-variance hedging
strategy
¯
h =(
¯
h
0
.
¯
h
1
) is determined via
¯
h
1
(t) =
¯
h
1
(t. H) +
¯
\(t)

¯
V(t −1) −
¯
V(0) −
t−1
¸
;=1
¯
h
1
(;)AX(;)

. (5.70)
which involves processes from the following crucial decomposition:
H =
¯
H(0) +
T
¸
;=1
¯
h
1
(;. H)AX(;) +
¯
L(T. H). (5.71)
where
¯
L(H) is a martingale (under the true probability measure P), and where
¯
L(H) in addition has the special property that also the product of
¯
L(H) and
(X−A) is a P-martingale. The process
¯
V is determined as follows:
¯
V(t) =
¯
E|H(t)].
Furthermore, the optimal initial capital (the approximation price) is in this
case given by
¯
V(0) =
¯
E|H] =H(0). Finally we mention that the variance of
5.7 Hedging integrated risks 193
the difference between the optimal investment strategy and the liability H is
given by
Var|N(H)] =E|(N(H))
2
] =
T
¸
t=1
E|(A
¯
L(t. H))
2
]
T
¸
;=t+1
(1−
¯
\(;)AA(;));
(5.72)
see Schweizer (1995, Theorem 4.4). This quantity can be interpreted as a
measure for the risk that cannot be hedged away. We show below how
Equation (5.72) appears as a safety loading under some alternative premium
calculation principles.
5.7.7 Mean-variance hedging for unit-linked contracts
This section gives the mean-variance strategy for the unit-linked contract,
Equation (5.46), within the market introduced in Section 5.7.1. Via some
tedious calculations, which can be found in Section A.4 of the Appendix, we
reach the following result.
Proposition 5.8 The approximation price for the unit-linked contract is
given by /
x T
¡
x
r(0. 1 ). The mean-variance hedging strategy is defined recur-
sively via
¯
h
1
(t) = Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1)
+
¯
\(t)

Y(t −1)
T−(t−1)
¡
x+(t−1)
r(t −1. 1)
−/
x T
¡
0
r(0. 1) −
t−1
¸
;=1
¯
h
1
(;)AX(;)

.
¯
h
0
(t) = /
x T
¡
x
r(0. 1) +
t
¸
;=1
¯
h
1
(;)AX(;) −
¯
h
1
(t)X(t).
We comment on this result. The optimal number
¯
h
1
(t) of stocks to be
held during the interval (t −1. t] is calculated as the number from the risk-
minimizing strategy with the addition of a correction term. This correction
term is defined recursively, and loosely speaking it compares the change
¯
V(t −1) −
¯
V(0) =Y(t −1)
T−(t−1)
¡
x+(t−1)
r(t −1. 1) −/
x T
¡
0
r(0. 1 )
in the fair price and the realized investment gains
t−1
¸
;=1
¯
h
1
(;)AX(;)
194 Integrated actuarial and financial valuation
for the period |0. t −1]. If the investment gains do not match the change
in
¯
V, the mean-variance strategy deviates from the risk-minimizing strategy.
In addition, we mention that the mean-variance strategy requires an initial
investment of /
x T
¡
x
r(0. 1 ) and that the strategy is self-financing.
The two criteria of risk minimization and mean-variance hedging are based
on a symmetric criterion that treats gains and losses in the same way (they
are to be avoided). Moreover, under these two criteria, the risk that cannot be
hedged via a self-financing strategy is not part of the premium. It might be
more natural if these quantities appeared in the form of a safety loading on
the fair premium. This is exactly the case in Section 5.7.8.
5.7.8 Indifference pricing and hedging
We start by considering a so-called mean-variance utility function given by
u
8
(Y) =E|Y ] −o(Var|Y])
8
. (5.73)
where Y is the wealth of the company at time T. (We refer to Dana (1999)
for a treatment of mean-variance utility functions.) We focus on the two situ-
ations where 8 =1¡2 and 8 =1, respectively. Equation (5.73) is considered
as a description of the life insurance company’s preferences. The classical
actuarial variance and standard deviation premium calculation principles can
be derived from these utility functions via a so-called indifference argument.
More precisely, we say that an insurance company with the utility function
given by Equation (5.73) prefers the pair (¡
8
. H) (i.e. selling the contingent
claim H and receiving the premium ¡
8
) to the pair (¡

8
. H

) if
u
8

8
−H) ≥u
8

8
−H

).
In particular, the pair (0, 0) corresponds to not selling any contingent claims
and not receiving any premium, and the insurance company is now said to
be indifferent (with respect to u
8
) between selling H at the price ¡
8
and not
selling H, provided that
u
8

8
−H) =u
8
(0) =0. (5.74)
i.e. provided that
¡
8
=−u
8
(−H) =E|H] +o(Var|H])
8
.
This shows that the classical actuarial premium calculation principles can
indeed be derived from the mean-variance utility functions given by Equa-
tion (5.73).
5.7 Hedging integrated risks 195
We now extend the indifference argument by including the insurance
company’s possibilities for investing in the financial markets. In particular,
this involves a more precise description of the company’s wealth at a future
time T. Assume that the company faces a liability H (payable at T) and that
the company has received the premium v
8
. Denote by V(0) the company’s
initial capital at time 0. If the insurance company follows an investment
strategy h with the initial investment V(0. h) =V(0), this leads to the wealth
at T as follows:
V(T. h) +v
8
−H =V(0) +v
8
+
T
¸
;=1
h
1
(;)AX(;) −H.
Alternatively, the insurance company might decide not to accept the liability
H and simply follow an investment strategy
¨
h with V(0.
¨
h) =V(0) and thus
obtain the following wealth:
V(T.
¨
h) =V(0) +
T
¸
;=1
¨
h
1
(;)AX(;).
One could say that the company is indifferent between these two different
possibilities if the premium v
8
is determined such that
u
8

V(T. h) +v
8
−H

=u
8

V(T.
¨
h)

.
However, we have not yet specified how the insurance company chooses the
investment strategies h and
¨
h, and each choice of strategies could potentially
lead to different premiums v
8
. A natural way to define the fair premium v
8
is instead to compare the utility associated with an optimal strategy in the
two cases where the company has accepted the liability and where it has not
accepted the liability. This leads to the following definition.
Definition 5.9 The indifference price v
8
(H) for H is defined by
sup
h:V(0.h)=V(0)
u
8
(V(T. h) +v
8
(H) −H)=
!
sup
¨
h:V(0.
¨
h)=V(0)
u
8
(V(T.
¨
h)).
The special case with no investments can, for example, be covered by
assuming that V(T. h) =V(0. h) for all investment strategies. In this case we
recover Equation (5.74) from this definition, and hence the classical actuarial
variance and standard deviation principles.
From Møller (2001b) and Schweizer (2001b), we have the following results
for the solution to the indifference pricing principles. Under the variance
196 Integrated actuarial and financial valuation
principle, i.e. the case where 8 =1, the fair premium is given by
v
1
(H) =
¯
E|H] +oVar|N(H)]. (5.75)
where N(H), defined in Equation (5.68), is the part of H that cannot be hedged
away. The investment strategy that maximizes the utility of the company is
given by
h
1
1
(t) =
¯
h
1
(t) +
1+Var|
¯
Z(T)]
2o
¯
8(t). (5.76)
where
¯
8(t) =
¯
\(t)
t
¸
;=1
(1−
¯
\(;)AX(;)).
and where
¯
Z(T) is defined by Equation (5.69). For the standard-deviation
principle, i.e. the case where 8 =1¡2, the fair premium is given by
v
1¡2
(H) =
¯
E|H] +¯o

Var|N(H)] (5.77)
if o
2
≥Var|
¯
Z(T)], where
¯o =o

1−
Var|
¯
Z(T)]
o
2
.
If o
2
-Var|
¯
Z(T)], then the u
1¡2
-indifference price is not defined. The optimal
strategy is well defined provided that o
2
> Var|
¯
Z(T)], and it is in this case
given by
h
1
2
(t) =
¯
h
1
(t) +
1+Var|
¯
Z(T)]
¯o

Var|N(H)]
¯
8(t). (5.78)
The principles given in Equations (5.75) and (5.77) are also called the financial
variance and standard deviation principles, respectively. We see that the two
fair premiums resemble the classical premium calculation principles in that
they consist of two terms: an expected value of H and a safety loading.
For the financial counterparts, the expected value of H is calculated under
a martingale measure, whereas the safety loading involves the variance of
that part of H which cannot be hedged away via a self-financing strategy.
Note, moreover, that this variance is calculated with respect to the original
probability measure P.
For extensions and further results on the financial variance and standard
deviation principles and their applications in insurance, see Møller (2002,
5.7 Hedging integrated risks 197
2003a, b). Indifference pricing for insurance contracts under exponential util-
ity functions has been considered by Becherer (2003). For a survey on utility
indifference pricing, see the forthcoming book edited by R. Carmona.

5.7.9 Indifference pricing for unit-linked contracts
The results from the previous sections show that the fair premiums and
optimal investment strategies are related to the results from mean-variance
hedging and risk minimization. Thus, all the hard work has already been
done, and it only remains to calculate Var|N(H)] given by Equation (5.72)
for the decomposition in Equation (5.63), i.e. with
¯
L(t. H) =
t
¸
;=1
r(;. 1)AM(;).
The variance of N(H) can be determined via calculations similar to the ones
used in Section 5.7.5; see Equations (5.64) and (5.65). In this way, we obtain
E|(A
¯
L(t. H))
2
] = E|(r(t. 1 ))
2
]E|AM(t)
2
]
= E|(r(t. 1 ))
2
] /
x T
¡
x T−t
¡
x+t
(1−
1
¡
x+(t−1)
). (5.79)
Thus,
Var|N(H)] =
T
¸
t=1

E|(r(t. 1))
2
]E|AM(t)
2
]

T
¸
;=t+1
(1−
¯
\(;)AA(;)). (5.80)
We see from Equation (5.79) that E|(A
¯
L(t. H))
2
] and E|AM(t)
2
] are pro-
portional to /
x
. This implies that the variance of N(H) is proportional to the
number /
x
of persons insured, whereas the standard deviation of N(H) is
proportional to the square root of /
x
.
We can now write up the main results as follows.
Proposition 5.10 The fair premium for the unit-linked contract under the
variance principle is given by
v
1
(H) =/
x T
¡
x
r(0. 1) +oVar|N(H)].

V. Henderson and D. Hobson. Utility indifference pricing – an overview, in R. Carmona
(ed.),
Indifference Pricing (to be published by Princeton University Press).
198 Integrated actuarial and financial valuation
where Var|N(H)] is given by Equation (5.80). The optimal strategy is deter-
mined by
h
1
1
(t) = Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1 )
+
¯
\(t)

Y(t −1)
T−(t−1)
¡
x+(t−1)
r(t −1. 1 )
−/
x T
¡
0
r(0. 1 ) −
t−1
¸
;=1
¯
h
1
(;)AX(;)

+
1+Var|
¯
Z(T)]
2o
¯
8(t).
We note the following. The fair premiumunder the financial variance princi-
ple is the sumof the premiumunder the criterion of mean-variance hedging and
a safety loading, which is proportional to /
x
. The optimal strategy consists of two
terms: the mean-variance strategy (the second and third lines) and a correction
term (the fourth line), where the latter does not depend on the liability H. This
last term is closely connected to the variance principle, and it is present even in
the case where the company has not accepted any liabilities.
For the standard deviation principle, we get a similar result, which, for
completeness, we formulate here.
Proposition 5.11 The fair premium for the unit-linked contract under the
standard deviation principle is given by
v
1
(H) =/
x T
¡
x
r(0. 1) +¯o

Var|N(H)].
where Var|N(H)] is given by Equation (5.80). The optimal strategy is deter-
mined by
h
1
2
(t) = Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1)
+
¯
\(t)

Y(t −1)
T−(t−1)
¡
x+(t−1)
r(t −1. 1)
−/
x T
¡
0
r(0. 1) −
t−1
¸
;=1
¯
h
1
(;)AX(;)

+
1+Var|
¯
Z(T)]
¯o

Var|N(H)]
¯
8(t).
5.7.10 Quantile hedging
An interesting alternative to the above treated principles is quantile hedging;
see, for example, Föllmer and Leukert (1999) and Föllmer and Schied (2002).
5.8 Traditional life insurance 199
Definition 5.12 (Quantile hedging) Given V(0) -sup
Q∈Q
E
Q
|H],
moximize P

V(0) +
T
¸
t=1
h
1
(t) AX(t) ≥H

over self-financing strategies h with V(0. h) =V(0).
The main idea of quantile hedging is the following. Assume that the
company has an initial capital V(0) and is interested in using this amount to
reduce the risk associated with the liability H. If the amount is insufficient to
super-replicate H, one could alternatively look for the strategy that maximizes
the probability that the terminal wealth V(T. h) exceeds the liability H.
The solution is well understood in the complete market case, where there
exists a unique equivalent martingale measure. To formulate this result, let
Q be the unique martingale measure. It follows from Föllmer and Leukert
(1999) that the solution is closely connected to finding an event
¯
A ∈ (T)
with
P(
¯
A) =max]P(A)A ∈ (T). E
Q
|H1
A
] ≤V(0)]. (5.81)
The main result basically says that quantile hedging for the liability H is
similar to performing super-hedging for the modified liability H1
¯
A
, which is
equal to H on the set
¯
A and zero otherwise.
The incomplete market case is less well understood. The main problem is
essentially that there are infinitely many martingale measures and that it is not
really clear which martingale measure one should choose in Equation (5.81)
in the general case. As a consequence, we cannot at this point give the quantile
hedging strategy for the unit-linked contract.
5.8 Traditional life insurance
In this chapter, we have focused on the analysis of unit-linked contracts. The
main reason for this choice is that these contracts provide a clearly specified
connection between benefits and returns on the underlying assets. Once the
company’s bonus and investment strategy has been specified as in Chapter 4,
the company’s liability associated with traditional, participating life insurance
contracts may also be viewed as an integrated contingent claim. Thus, one
can, in principle, apply the same methods for the treatment of this integrated
claim if it is possible to determine fairly explicit expressions for the liability.
6
Surplus-linked life insurance
6.1 Introduction
We consider in this chapter the general type of life insurance where premiums
and benefits are calculated provisionally at issuance of the policy and later
determined according to the performance of the insurance contract or com-
pany. The determination of premiums and benefits can take various forms
depending on the type of contract. Examples are various types of participating
life insurance (in some countries called with-profit life insurance) and various
types of pension funding.
The determination of premiums and benefits is based on payment of div-
idends, which in general may be positive or negative, from the insurance
company to the policy holder. It is important to distinguish between two
aspects of the determination: the dividend plan and the bonus plan. The divi-
dend plan is the plan for allocation of dividends. However, often the dividends
are not paid out immediately in cash but are converted into a stream of future
payments. The bonus plan is the plan for how the dividends are eventually
turned into payments.
In Steffensen (2000), a framework of securitization is developed where
reserves are no longer defined as expected present values but as market
prices of streams of payments (which, however, happen to be expressible
as expected present values under adjusted measures). An insurance contract
is defined as a stream of payments linked to dynamic indices, covering a
wide range of insurance contracts including various forms of unit-linked
contracts. Securitization is one way of dealing simultaneously with the risk
in the insurance policy and the risk in the financial market. It is built on the
consideration of the stream of payments stipulated in an insurance contract as
a dynamically traded object on the financial market. The insurance company
is then considered as a participant in this market that has to adapt prices and
200
6.1 Introduction 201
strategies to the market conditions. The framework of securitization can be
argued for in two different ways.
Firstly, the insurance contracts are indeed in certain ways dynamically
traded securities. The insurance company and the policy holder exchange
payment streams in a competitive market. The policy holder can typically
surrender their contract and hereby give up their position in the security.
The insurance company can primarily take the opposite position by reselling
(parts of the) risk aggregated in portfolios or sub-portfolios. This happens
when the insurance company buys reinsurance cover or sells, for example,
non-core business lines to competing insurance companies. Indeed, market
prices of risk may be hidden, blurred, ambiguous and hard to obtain due to
a non-effective, non-liquid and maybe even obscure market place. They may
still be the best source of “best market value estimates,” however.
Secondly, even if the market information is found to be inadequate in this
respect, the supervisory authorities may implement accounting and solvency
standards which are built on this pattern of thinking. By this we mean that such
standards could, for example, dictate certain prices of risk for calculation of
entries. To this extent the idea of securitization is rather a point of view of the
regulators, and one could of course discuss whether such prices of risk should
be spoken of as market prices. For a general description of securitization in
life insurance, see Cowley and Cummins (2005).
In this chapter we construct a general life and pension insurance contract
where we link dividends to a certain notion of surplus within the framework
developed in Steffensen (2000). Working with general surplus-linked pay-
ments in participating life insurance and pension funding, we go beyond the
traditional setup of payments in the existing literature on the emergence of
surplus and dividends. However, surplus-linked payments offer a number of
appealing setups. The study of such surplus-linked insurance has a three-fold
motivation, as we see in the following.
Firstly, it represents a new product, combining properties of participating
life insurance, pension funding and unit-linked insurance. In fact, a major
Danish life insurance company introduced in 2002 a new product under the
name tidspension (time pension), which can be analyzed within the framework
presented in this chapter.
Secondly, the Danish legislation shows a trend towards a formalization
of the participation element of participating life insurance contracts. This
trend can be expected to continue world-wide when international rules for
market valuation in participating life insurance and pension funding are con-
cretized; see, for example, Jørgensen (2004) for the status of accounting
standards and fair valuation. The present chapter serves as an example of
such a formalization.
202 Surplus-linked life insurance
Thirdly, it seems to represent a good imitation of the behavior of managers.
As such, it can be used as a management tool as well as a market analysis
tool.
Subjugating life and pension insurance to the market conditions, the appro-
priate tool seems to be mathematical finance or, more specifically, contin-
gent claims analysis. Option pricing theory was introduced as a tool for the
analysis and management of unit-linked insurance in the 1970s (see Brennan
and Schwartz (1976) and references in Aase and Persson (1994)). The con-
sideration of schemes in pension funding as options also goes back to the
1970s (see, for example, Sharpe (1976) and references in Blake (1998)). In
participating life insurance, however, contingent claims analysis as a tool for
analysis and management has been long in coming and was, to our knowl-
edge, introduced by Briys and de Varenne (1994). Since then, the framework
of Briys and de Varenne (1994) has been developed further and generalized
substantially by several financial economists: Grosen and Jørgensen (2000),
Hansen and Miltersen (2002) and Miltersen and Persson (2003) are examples
of some more recent references.
Contingent claims analysis of participating life insurance has been long
in coming mainly for one reason. The link between the payments and the
performance of the company in participating life insurance is often laid down
by statute so vaguely that it may seem unreasonable to consider dividends
as contractual. Working in a framework of securitization, our main objection
to this argument is, of course, that the insurance business, and hereby the
participation in the performance, takes place in a competitive market. Thus,
the insurance company is forced to adapt, for example, its plans for the
determination of payments to the market conditions. This objection is, at the
same time, the primary argument for applying contingent claims analysis to
life and pension insurance at all.
In Grosen and Jørgensen (2000), Hansen and Miltersen (2002) and
Miltersen and Persson (2003), the main focus has been on the financial
elements of the insurance contracts. In contrast, other authors have con-
tributed to the understanding of the actuarial notion of surplus, including
possible insurance risk, and its redistribution in terms of dividends and
bonus; see Norberg (1999) and Ramlau-Hansen (1991). Norberg (1999) works
with a stochastic economic–demographic environment, though with only one
investment opportunity. Steffensen (2001) approaches the actuarial notion
of surplus using the methods and terminology of mathematical finance, and
hereby links classical (financial) modeling of financial risk and classical
(actuarial) modeling of insurance risk. This chapter is very much based on
Steffensen (2006b), in which the notion of surplus-linked life insurance is
introduced.
6.2 The insurance contract 203
The chapter is structured as follows. In Section 6.2 we construct the
framework for the insurance contract. In Section 6.3 the notions of surplus
are studied and related to previous literature. In Section 6.4 we consider
general surplus-linked dividend plans, and the particular situation of a linear
link is studied in Section 6.5. Section 6.6 works out a framework for bonus.
In section 6.7 we link dividends to the surplus and a certain bonus index, and
the particular situation of a linear link is then studied in Section 6.8. At the
end of each section we have added a remark which explains how the present
section connects to parts of the material presented in Chapters 2 and 4.
6.2 The insurance contract
We take as given a probability space (D. . P). On the probability space is
defined a process Z =(Z(t))
0≤t≤n
taking values in a finite set =]0. . . . . 1]
of possible states and starting in state 0 at time 0. We define the 1-dimensional
counting process N =

N
k

k∈
as follows:
N
k
(t) =# ]s s ∈ (0. t] . Z(s−) =k. Z(s) =k] .
counting the number of jumps into state k until time t. Assume that there exist
deterministic and sufficiently regular functions µ
;k
(t), ;. k ∈ , such that N
k
admits the stochastic intensity process

µ
Z(t)k
(t)

0≤t≤n
for k ∈ , i.e.
M
k
(t) =N
k
(t) −

t
0
µ
Z(s)k
(s) ds
constitutes a martingale for k ∈ . Then Z is a Markov process. The reader
should think of Z as a policy state of a life insurance contract; see Hoem
(1969) for a motivation for the setup. We let
N
=
Z
denote the filtration
generated by Z formalizing information about Z, i.e.

N
(t) =u (Z(s) . 0 ≤s ≤t) .
We emphasize that the intensity process

µ
Z(t)k
(t)

0≤t≤n
is adapted to
N
such that there is no demographic risk.
Also defined on the probability space is a standard Brownian motion W.
We consider a financial market with two assets which are continuously
204 Surplus-linked life insurance
traded. The market prices are described by the following stochastic differential
equation:
dS
0
(t) =rS
0
(t) dt.
S
0
(0) =1.
dS
1
(t) =rS
1
(t) dt +uS
1
(t) dW (t) .
S
1
(0) =s
0
.
where r and u are constants. This is the classical Black–Scholes market. We
let
W
=
S
1
denote the filtration generated by S
1
formalizing information
about S
1
, i.e.

W
(t) =u

S
1
(s) . 0 ≤s ≤t

.
and we assume that
=
N

W
.
In the dynamics of S
1
above, we take the short rate of interest r to be the
expected infinitesimal return and hereby take P to be the valuation measure
not the physical measure. Throughout the chapter we are only interested in
valuation, and it is convenient then to specify all dynamics directly under the
valuation measure. The same goes for dynamics of Z above in the sense that
we take M
k
to be a martingale under the valuation measure. Thus, we think
of having fixed a market valuation intensity. Whether this intensity has come
to our knowledge from prices of other insurance contracts in the market, or
from some alternative determination of market attitudes towards insurance
(policy) risk, is not important. We follow Steffensen (2000) and speak of
conditional expected values under P as arbitrage-free values. The difference
in comparison with Steffensen (2000) is that here we have specified the
dynamics directly under the valuation measure in the first place.
We consider a standard life insurance payment process; i.e. the accumulated
contractual benefits less premiums are described as follows:
dB(t) =dB
Z(t)
(t) +
¸
k∈
l
Z(t−)k
(t) dN
k
(t) .
dB
;
(t) =l
;
(t) dt +AB
;
(t) . ; ∈ .
where l
;
(t), AB
;
(t) and l
;k
(t) are deterministic and sufficiently regular
functions. The sum AB
;
(t) represents a lump sum payment at the deter-
ministic time point t if Z(t) = ;. The set of deterministic time points with
discontinuities in B is given by =
¸
t
0
. t
1
. . . . . t
q
¸
.
6.2 The insurance contract 205
We consider the deterministic basis (r. µ), which specifies an interest rate
and a set of transition intensities. On this basis we introduce the statewise
reserves for ; ∈ and for t ∈ |0. n] as follows:
V
B;
(t) =E
t.;
¸

n
t
e

s
t
r
dB(s)
¸
. (6.1)
where E
t.;
denotes the expectation conditioned on Z(t) = ; and where

n
t
=

(t.n]
.
Since the expectation in Equation (6.1) is taken under the market valuation
measure, we speak of Equation (6.1) as the market value or arbitrage-free
value of the payment process B. Financial mathematics teaches us that such
a market value can always be written as such a conditional expected present
value with an appropriate specification of underlying dynamics. The basis
(r. µ) is hereafter spoken of as the market basis.
There is a deep connection between conditional expected values as in Equa-
tion (6.1) and solutions to corresponding systems of deterministic differential
equations, henceforth spoken of simply as differential equations. This con-
nection is used throughout the chapter, with a reference to Steffensen (2000)
where this connection is proved in sufficient generality for all our applica-
tions. The differential equation for V
B;
(t) is, however, very well known from
various other references, since this is the general Thiele differential equation.
However, for consistency, we also follow Steffensen (2000) for the following
ordinary differential equation for V
B;
:
V
B;
t
(t) =rV
B;
(t) −l
;
(t) −
¸
k:k=;
µ
;k
(t) R
B;k
(t) . t ∈ |0. n] \. (6.2a)
V
B;
(t−) =AB
;
(t) +V
B;
(t) . t ∈ . (6.2b)
R
B;k
(t) =l
;k
(t) +V
Bk
(t) −V
B;
(t) . (6.2c)
where a subscript denotes (partial) differentiation, i.e. V
t
(t) =(o¡ot) V (t). In
Equation (6.2a) we have the ordinary differential equation valid for V
B;
(t) at
all times that are outside . Differentiability follows from sufficient regularity
of the coefficients of Equation (6.2a). This differential equation involves a
so-called sum at risk R
B;k
(t) given in Equation (6.2c). At the time points
of discontinuity, i.e. inside , the pasting condition, Equation (6.2b), holds.
The terminal condition of the differential equation (6.2) is, by definition,
V
B;
(n) =0. Throughout, we skip the specification of the terminal condition,
which will be evident and given by definition.
An important quantity in life insurance is the value of future payments
on a so-called technical basis (r

. µ

). Which technical basis to use depends
206 Surplus-linked life insurance
on the context or the purpose of the valuation. Given a technical basis, the
technical reserve,
V
B∗;
(t) =E

t.;
¸

n
t
e

s
t
r

dB(s)
¸
.
fulfils the following ordinary differential equation:
V
B∗;
t
(t) =r

V
B∗;
(t) −l
;
(t) −
¸
k:k=;
µ
∗;k
(t) R
B∗;k
(t) . t ∈ |0. n] \. (6.3)
V
B∗;
(t−) =AB
;
(t) +V
B∗;
(t) . t ∈ .
R
B∗;k
(t) =l
;k
(t) +V
B∗k
(t) −V
B∗;
(t) .
Here, E

denotes the expectation with respect to µ

.
A special purpose of a special technical basis is the calculation of the
payment process B at time 0. The technical basis used for this purpose is
called the first order basis. The payment process B is calculated in accor-
dance with the so-called equivalence relation performed under the first order
basis. If (r

. µ

) is the first order basis, the equivalence relation is as
follows:
V
B∗0
(0−) =0. (6.4)
We speak of the payment process which results from this equation as the
first order payment process. Thus, the first order payment process is simply
the payments that are specified in the contract at the time of issue. The
equivalence relation given in Equation (6.4) implies a balance between the
first order expected present value of the first order premiums and the first
order expected present value of the first order benefits.
If the first order basis is chosen as the technical basis, the equivalence
relation Equation (6.4) is fulfilled. This relation can then be viewed as an
initial condition for V
B∗0
. With this initial condition and the differential equa-
tion in Equation (6.3) for the state ; =0, one can write down a retrospective
(calculated from 0− to t) solution V
B∗0
(t) in terms of transition probabilities
of Z. This had been spoken of as a retrospective reserve until Norberg (1991)
pointed out that this was nothing but a retrospective formula for the prospec-
tive value V
B∗0
(t) based on the equivalence relation for V
B∗0
. Norberg (1991)
then introduced a different notion of retrospective reserve based on a present
valuation of past payments.
In order to rectify the possible non-equivalence of the first order payments
under the market basis in the sense of V
B0
(0−) =0, the insurance company
6.2 The insurance contract 207
adds dividends to the first order payments. We denote by D the process of
accumulated dividend payments, and we assume that this is described by
dD(t) =dD
Z(t)
(t) +
¸
k∈
o
Z(t−)k
(t) dN
k
(t) . (6.5)
dD
;
(t) =o
;
(t) dt +AD
;
(t) . ; ∈ .
where o
;
(t), AD
;
(t) and o
;k
(t) are, in general, -adapted stochastic pro-
cesses. Note that the coefficients of the payment process D are substantially
different from the coefficients of the payment process B. The dividend pay-
ments are allowed to depend, in general, on the full history and on the history
of capital gains and losses in particular.
We now constrain dividend processes in the form of Equation (6.5) to
be fair in some sense. Norberg (1999, 2001) studies fairness constraints on
dividends as well in an economic–demographic environment driven by a
stochastic market basis. The stochastic interest rate in the stochastic market
basis represents the only investment opportunity. The approach by Norberg
(1999, 2001) is to constrain dividends to attain ultimate fairness in the sense of
equivalence conditional on the economic–demographic history. Formalizing
by
(r.µ)
this history, the fairness constraint by Norberg (1999, 2001) is as
follows:
E
¸

n
0−
e

t
0
r
d (B+D) (t)

(r.µ)
(n)
¸
=0. (6.6)
Note that this fairness constraint prevents the insurance company from taking
ultimate economic–demographic risks.
In this chapter we have no demographic uncertainty, while economic uncer-
tainty is contained in
S
1
. Thus, a direct translation of Equation (6.6) to our
situation is as follows:
E
¸

n
0−
e

t
0
r
d (B+D) (t)

S
1
(n)
¸
=0. (6.7)
From a theoretical point of view, this could work very well as a fairness
constraint. In practice, however, the insurance companies do work with div-
idends which imply ultimate financial risk, indicating that Equation (6.7) is
too restrictive. This behavior may be demanded by policy holders. This does
not necessarily mean that Equation (6.6) can be said to be too restrictive,
though. Recall that Equation (6.6) is introduced in a fundamentally different
economic–demographic environment with locally risk-free capital gains.
The question is: which fairness criterion should replace Equation (6.7)
(and Equation (6.6)) such that the insurance companies are allowed to take
208 Surplus-linked life insurance
financial risks? From mathematical finance, we inherit the following
criterion:
E
¸

n
0−
e

t
0
r
d (B+D) (t)
¸
=0. (6.8)
This leads to what in Steffensen (2000) is spoken of as an arbitrage-free
insurance contract. The fundamental idea is that the constraint given in Equa-
tion (6.8) prevents arbitrage opportunities in an (artificial) market where
attitudes towards risk conform with the valuation measure P. We refer the
reader to Steffensen (2000) for further motivation of Equation (6.8) and mat-
ters arising from it.
It is clear, by the tower property, that Equation (6.8) is less restrictive than
Equation (6.7) since any dividend process which is fair according to Equa-
tion (6.7) is also fair according to Equation (6.8). Since Equation (6.8) says
much less about the dividends than Equation (6.7) does, the next important
step is to specify further the dividends. Since D is, in general, -adapted, it
is in a sense based on experience, although our experience basis differs
substantially from the experience basis presented in Norberg (1999, 2001), i.e.

(r.µ)
. The idea of the present chapter is to suggest how the dividends could
be based on the experience in order to construct new tractable and relevant
insurance products and to imitate (the behavior of managers of) old insurance
products. This chapter presents one family of dividend specifications based
on this idea.
A major drawback of going from Equation (6.6) to Equation (6.7) and from
Equation (6.7) to Equation (6.8) is that realistic modeling of S =

S
0
. S
1

and
Z are required. It is doubtful whether one can make realistic assumptions in
the long term perspective of a life insurance contract. Both the Black–Scholes
market and the Markovian Z applied here seem far too simple. We should
here emphasize the purpose of this chapter. It serves to present the concept of
surplus-linked life insurance payments and demonstrate the partial differential
equation methodology for calculation of reserves based on this concept. It
does not provide new realistic models for long term financial markets and
stochastic intensities.
The idea of surplus-linked life insurance payments generalizes directly
to any financial market, and the partial differential equation methodology
applies to a wide range of Markovian financial and intensity models. Such
markets include popular and well known Gaussian interest rate and volatility
models and less popular and less well known Gaussian intensity models. The
generalizations follow from appropriate textbooks on mathematical finance
6.2 The insurance contract 209
(see Björk (2004), for example) and insurance literature proposing stochastic
intensity models (see, for example, Steffensen (2000)).
Remark 6.1 Consider an endowment insurance in the survival model illus-
trated in Figure 6.1. We can skip the superfluous state specification in this
model, and we denote the death counting process by N, the transition intensity
by µ, the statewise reserve corresponding to state 0 by V
B
and the sum at
risk corresponding to transition from 0 to 1 by R
B
. Denote by l
ad
the death
sum paid out upon death, by l
a
(0) the lump sum paid out upon survival until
time n, and by r the continuous level premium paid as long as the insured
is alive. Then
dB(t) =dB
Z(t)
(t) +l
ad
dN (t)
dB
0
(t) =−rdt +l
a
(0) de
n
(t) .
dB
1
(t) =0.
where e
n
(t) =1|t ≥n] indicates that t exceeds n. Noting the general terminal
condition V
B
(n) =0, Equation (6.2) reduces to the following:
V
B
t
(t) =rV
B
(t) +r−µ(t) R
B
(t) .
V
B
(n−) =l
a
(0) .
R
B
(t) =l
ad
−V
B
(t) .
This is exactly the differential equation for V
g
in Equation (2.23) where we
put t = 0 (and then skip the t argument) since we are here valuating the
payments guaranteed at time 0.
Correspondingly, Equation (6.3) reduces to the following:
V
B∗
t
(t) =r

V
B∗
(t) +r−µ

(t) R
B∗
(t) .
V
B∗
(n−) =l
a
(0) .
R
B∗
(t) =l
ad
−V
B∗
(t) .
But this is exactly the differential equation for V

in Equation (2.10) where,
again, we put t =0 (and then skip the t argument).
0
alive
1
dead
Figure 6.1. Survival model.
210 Surplus-linked life insurance
6.3 Surplus
In Section 6.2 we introduced the process of dividends. In this section we
study a process which, at the same time, is the source of dividends and, in the
forthcoming sections, also determines the dividend payments. This process is
taken to be a wealth process, denoted by X, with income in the form of a
contribution process, which we call C, and with consumption in form of the
dividend process D. The wealth process is invested in the market described
in Section 6.2. The dynamics of this wealth process determines the results in
the following sections.
Usually one would speak of the source of dividends as some kind of
surplus. The dynamics of the wealth process introduced below is so gen-
eral that it includes various notions of surplus suggested in the literature.
We choose to speak of the wealth process simply as the surplus. After
having introduced the dynamics of the surplus below, we go through a
series of more or less congruent classical surplus definitions which are all
special cases.
We consider a surplus X with accumulated income process C and accumu-
lated consumption process D. We take u
Z(t)
(t. X(t)) ¡u to be the proportion
of the surplus invested in S
1
, where u
;
(t. x) is a deterministic and suffi-
ciently regular function; i.e. if the number of stocks held at time t is denoted
by r
1
(t), then
u
Z(t)
(t. X(t)) X(t) =ur
1
(t) S
1
(t) .
Recall that u is the volatility of S
1
. Then the surplus satisfies the following
stochastic differential equation:
dX(t) =rX(t) dt +u
Z(t)
(t. X(t)) X(t) dW (t) +d (C−D) (t) . (6.9)
X(0−) =0.
We assume that the income process C, also spoken of as the contribution
process, is described as follows:
dC(t) =dC
Z(t)
(t) +
¸
k∈
c
Z(t−)k
(t) dN
k
(t) . (6.10)
dC
;
(t) =c
;
(t) dt +AC
;
(t) . ; ∈ .
where c
;
(t), AC
;
(t) and c
;k
(t) are deterministic and sufficiently regular
functions. For the moment, we have no particular form of these coefficients
in mind; this is just a specification of the structure of contributions. Note that
this structure is similar to the structure of the payment process B.
6.3 Surplus 211
This is basically what the reader needs to know about the surplus in the
succeeding sections. However, in the rest of this section we relate these
surplus dynamics to particular notions of surplus, including the traditional
ones. These relations serve as our motivation for working with a surplus in
the form given in Equation (6.9).
We first introduce the basic notion of an individual surplus. An impor-
tant part of the individual surplus is the cash balance of past payments
including capital gains, which we denote by U. The dynamics of U is
given by
dU (t) =rU (t) dt +r
1
(t) S
1
(t) dW (t) −d (B+D) (t) .
U (0−) =0.
The individual surplus is now defined as the excess of this cash balance over
the technical reserve, i.e.
X(t) =U (t) −V
B∗Z(t)
(t) . (6.11)
X(0−) =0.
We have assumed above that u
Z(t)
(t. X(t)) ¡u is the proportion of the X
invested in S
1
. This means that from the cash balance U (t) = X(t) +
V
B∗Z(t)
(t), the amount

u
Z(t)
(t. X(t)) ¡u

X(t) is invested in S
1
such that the
capital gains from this investment become

u
Z(t)
(t. X(t)) ¡u

X(t) dS
1
(t) ¡
S
1
(t) with a diffusion part u
Z(t)
(t. X(t)) X(t) dW (t). The residual amount,
given by
U (t) −

u
Z(t)
(t. X(t)) ¡u

X(t) =V
B∗Z(t)
(t) +X(t)
×

1−

u
Z(t)
(t. X(t)) ¡u

.
is invested in S
0
. The cash balance is then also seen to follow the dynamics:
dU (t) =rU (t) dt +u
Z(t)
(t. X(t)) X(t) dW (t) −d (B+D) (t) .
U (0−) =0.
By Ito’s formula, we can calculate the dynamics of the individual surplus as
follows:
dX(t) =rU (t) dt +u
Z(t)
(t. X(t)) X(t) dW (t) −d (B+D) (t) (6.12)
−V
B∗Z(t)
t
(t) dt −

V
B∗Z(t)
(t) −V
B∗Z(t)
(t−)


¸
k∈

V
B∗k
(t) −V
B∗Z(t−)
(t−)

dN
k
(t) .
212 Surplus-linked life insurance
Now we can plug in the dynamics of B, D and V

. After some rearrangements,
one finally arrives at the dynamics given in Equation (6.9), with the following
specification of the coefficients of C:
c
;
(t) =(r −r

) V
B∗;
(t) +
¸
k:k=;

µ
∗;k
(t) −µ
;k
(t)

R
B∗;k
(t) (6.13)

¸
k:k=;
c
;k
(t) µ
;k
(t) .
c
;k
(t) =−R
B∗;k
(t) .
AC
;
(t) =
¸
−AB
;
(t) −V
B∗;
(t) . t =0.
0. t >0.
(6.14)
The individual surplus shows an application of a technical basis different
from the calculation of first order payments. The fact that we allow, in gen-
eral, the technical basis in the definition of individual surplus to be different
from the first order basis is indeed a generalization in comparison with other
introductions of surplus in the literature (see below). Note the consequence
that AC
0
(0) may be different from zero if the first order basis is not used as
the technical basis in the surplus. See Norberg and Steffensen (2005) for a
detailed study of the connection between a stochastic differential equation of
the type given in Equation (6.12) and its solution.
The individual surplus introduced here relates to the notions of surplus
introduced by Ramlau-Hansen (1988, 1991) and later by Norberg (1999,
2001). In general, Ramlau-Hansen (1988, 1991) and Norberg (1999, 2001)
work with only one investment opportunity, earning interest at a general
rate r, and do not account for the dividends on the cash balance. Furthermore,
they work with the first order basis as technical basis so that AC
0
(0) =0 by
the principle of equivalence. The notions of surplus introduced by Ramlau-
Hansen (1988, 1991) and Norberg (1999, 2001) can all be written in the
following form:
dX(t) =rX(t) dt +dC(t) .
X(0) =0.
Particular notions of surplus then correspond to particular specifications of
the coefficients of C(t).
The surplus introduced by Ramlau-Hansen (1988, 1991) corresponds to the
specification in Equation (6.13) where the market basis (r. µ) is deterministic,
as in this chapter. The individual surplus introduced by Norberg (1999, 2001)
also corresponds to the specification in Equation (6.13), but with the market
6.3 Surplus 213
basis being stochastic. The mean surplus introduced by Norberg (1999, 2001)
is obtained by then averaging away insurance risk. The dynamics of the mean
surplus is given by the following specification of the coefficients of C:
c
;
(t) =
¸
;
¡
0;
(0. t)

(r −r

) V
B∗;
(t) +
¸
k:k=;

µ
∗;k
(t) −µ
;k
(t)

R
B∗;k
(t)

.
c
;k
(t) =0.
where we note that c
;
(t) does not actually depend on ;.
Here we introduce what we call the systematic surplus. This is obtained by
starting out with the individual surplus and then disregarding the martingale
part of C. The systematic surplus then follows the dynamics in Equation (6.9),
with coefficients of C given by
c
;
(t) =(r −r

) V
B∗;
(t) +
¸
k:k=;

µ
∗;k
(t) −µ
;k
(t)

R
B∗;k
(t) .
c
;k
(t) =0.
AC
;
(t) =
¸
−AB
;
(t) −V
B∗;
(t) . t =0.
0. t >0.
It is important to note that unsystematic financial risk is fully present in
the systematic surplus as it was in the individual surplus. Thus, the term
“systematic” only applies to the insurance risk, so to speak. It is also
important to understand the difference between our systematic surplus and
the mean surplus introduced by Norberg (1999, 2001). Indeed, by aver-
aging away insurance risk in the individual surplus, the martingale part
vanishes. In addition, however, the insurance risk in the systematic part
of the individual surplus is averaged away. The systematic surplus intro-
duced here, where the martingale part of C is disregarded while the system-
atic part of C is fully present, seems to be the most important concept in
practice.
Norberg (1999, 2001) introduces a dividend reserve that only accounts for
the systematic part of the individual surplus, but which accounts for the divi-
dends. The dividend reserve based on the individual surplus contributions dif-
fers from our systematic surplus only by the economic/financial–demographic
model.
We remind the reader that in the rest of the chapter we simply work with
the general form, Equation (6.9). The reader can then pick their favorite notion
of surplus and plug in the coefficients of the contribution process for further
specialization.
214 Surplus-linked life insurance
Remark 6.2 Consider again the endowment insurance from Remark 6.1.
Assume first that the only investment possibility available is at the constant
interest rate r. Assume that dividends are paid out at rate o. Then the dynamics
of the mean surplus for this contract are given by, skipping the superfluous
state specification of the surplus contribution,
dX(t) =rX(t) dt +¡
00
(0. t) (c (t) −o(t)) dt
=rX(t) dt +e

t
0
µ
(c (t) −o(t)) dt.
X(0) =0.
c (t) = (r −r

) V
B∗
(t) +(µ

(t) −µ(t)) R
B∗
(t) .
This should be compared with the undistributed reserve introduced in Equa-
tion (2.15) with the following dynamics:
dX
(2.15)
(t) =(r +µ(t)) X
(2.15)
(t) dt +(c (t) −o(t)) dt.
X
(2.15)
(0) =0.
There are two differences which need special attention. Since in this chapter
we have not yet introduced the payment of additional benefits for the divi-
dends, the contribution rate c is generated from the reserve V
B∗
(t) and the
sum at risk R
B∗
(t) that connects exclusively to the payments guaranteed at
time 0. The reserve and the sum at risk appearing in Equation (2.16) are
different since they concern all payments guaranteed in the past, not only
at time 0. In the context of this chapter, we introduce additional benefits
in Sections 6.6–6.8, and the precise connection between surplus here and
the undistributed reserve in Section 2.2.3 turns up in Remark 6.6. Further-
more, the mortality intensity comes into play differently. The connection is that
the undistributed reserve introduced in Equation (2.15) is actually defined as
the mean surplus per expected survivor, i.e.
X
(2.15)
(t) =
X(t)
E|I (t)]
=X(t) e

t
0
µ
.
where I (t) = 1|Z(t) =0] indicates survival until time t. This relation is
easily established by differentiation:
dX
(2.15)
(t) =e

t
0
µ
dX(t) +X(t) de

t
0
µ
=(r +µ(t)) X
(2.15)
(t) dt +(c (t) −o(t)) dt.
6.4 Surplus-linked dividends 215
6.4 Surplus-linked dividends
In this section we study a particular dividend process, namely a dividend
process where the dividend payments are linked to the surplus defined in
Section 6.3. We formalize this by defining the processes o
;
(t), o
;k
(t) and
AD
;
(t) as functions of X(t):
o
;
(t) =o
;
(t. X(t)) .
o
;k
(t) =o
;k
(t. X(t)) . (6.15)
AD
;
(t) =AD
;
(t. X(t)) .
where we, by abuse of notation, use the same letters for the processes and the
functions.
The purpose of this study is two-fold. Firstly, we suggest Equation (6.15)
for valuation, management and supervision in participating life insurance and
pension funding. In both participating life insurance and pension funding,
competition and supervision sees to it that a (positive) surplus generated
by policy holders is redistributed in terms of dividends. Traditionally, the
legislative formulation of this redistribution leaves degrees of freedom to
the insurance company. However, recently insurance companies and supervi-
sory authorities have focused more on measuring redistribution fairness for
valuation and management. For this purpose, a formalization of practice is
needed which balances practical complexity against mathematical tractability.
In Equation (6.15), we suggest such a formalization.
Secondly, we suggest the formalization in Equation (6.15) for general
product design in life and pension insurance. Traditional participating life
insurance has been criticized for lacking transparency. One could meet this
criticism by replacing a (vague) formulation of surplus participation con-
straints in the legislative environment by a (non-vague) formulation in the
contracts. Then Equation (6.15) makes up such a non-ambiguous formulation
which captures the main characteristics of present practice. In fact, a major
Danish life insurance company has recently introduced a life insurance prod-
uct where dividends are specified as linear functions of the surplus. Such a
construction of dividends is studied in detail in Section 6.5.
We refer to Steffensen (2000, 2001) for the deterministic differential equa-
tion that characterizes the reserve V under the specification in Equation (6.15).
In Steffensen (2000, 2001) the state process on which all payments depend
is required to be Markovian. Here the state process on which all payments
depend is (Z(t) . X(t)). Informally, (Z(t) . X(t)) is seen to be Markovian
since Z is Markovian and since all coefficients of the stochastic differential
216 Surplus-linked life insurance
equation for X depend on (Z(t) . X(t)) only; see Equations (6.9), (6.10),
(6.5) and (6.15). From Steffensen (2000, 2001), we see that the deterministic
function given by
V
;
(t. x) =E
¸

n
t
e

s
t
r
d (B+D) (s)

Z(t) =;. X(t) =x
¸
is fully characterized by the following differential equation:
V
;
t
(t. x) =rV
;
(t. x) −l
;
(t) −o
;
(t. x) −
¸
k:k=;
R
;k
(t. x) µ
;k
(t)
−V
;
x
(t. x)

rx+c
;
(t) −o
;
(t. x)


1
2
u
;
(t. x)
2
x
2
V
;
xx
(t. x) . t ∈ |0. n] \. (6.16a)
V
;
(t−. x) =AB
;
(t) +AD
;
(t. x) +V
;

t. x
;

. t ∈ . (6.16b)
R
;k
(t. x) =l
;k
(t) +o
;k
(t. x) +V
k

t. x
;k

−V
;
(t. x) . (6.16c)
x
;k
=x+c
;k
(t) −o
;k
(t. x) .
x
;
=x+AC
;
(t) −AD
;
(t. x) .
The differential equations given in Equations (6.16a–c) are similar to Equa-
tion (6.2). In Equation (6.16a) we have the partial differential equation valid
for V
;
(t. x) at all times that are outside . Differentiability follows from
sufficient regularity of the coefficients of Equation (6.16a). This differen-
tial equation involves the sum at risk R
;k
(t. x) given in Equation (6.16c).
At the time points of discontinuity, i.e. inside , the pasting condition in
Equation (6.16b) holds.
The differential equation can be used to calculate a set of fair functions,
Equations (6.15), leading to equivalence in the sense of
V
0
(0−. 0) =0.
This corresponds to the equivalence relation, Equation (6.8). In general, the
differential equation (6.16a) must be approached by numerical methods. In
special cases we can, however, reduce the partial differential equation to a set
of ordinary differential equations. Such a special case is studied in Section 6.5.
6.5 Dividends linear in surplus 217
Remark 6.3 Assume that we have a purely financial contract such that the
process Z is not needed. Then we can also disregard the state specification
of all quantities and the risk premia in Equation (6.16a) vanish. Assume that
there are no lump sum payments at deterministic points in time except for
at termination, i.e. = ]n], that the terminal benefit is given by l (0) and
that the continuous payment is a constant premium rate r, i.e. l (t) = −r.
Noting the general terminal condition V (n. x) =0, Equations (6.16) reduce
to the following:
V
t
(t. x) = rV (t. x) +r−o(t. x)
−V
x
(t. x) (rx+c (t) −o(t. x))

1
2
u (t. x)
2
x
2
V
xx
(t. x) .
V (n−. x) = l (0) +AD(n. x) .
This partial differential equation can now be compared with the partial
differential equations from Sections 4.5.1 and 4.5.2. If no dividends are
paid out before termination of the contract, i.e. o(t. x) = 0, we recognize
Equation (4.44) where the proportion of X invested in S
1
(here described by
u (t. x) ¡u) is denoted by O(t. x) and where AD(n. x) = 4(x). If instead
dividends are paid out continuously during the term of the contract, we
recognize Equation (4.45) where o(t. x) =4(t. x).
6.5 Dividends linear in surplus
In this section we specify further the dividend formalization suggested in
Equations (6.15). We derive a semi-explicit solution to Equations (6.16a–c)
in the case where the dividend payments are linear functions of the
surplus, i.e.
o
;
(t) =q
;
(t) X(t) .
o
;k
(t) =q
;k
(t) X(t) . (6.17)
AD
;
(t) =AQ
;
(t) X(t) .
where q
;
(t), q
;k
(t) and AQ
;
(t) are deterministic and sufficiently regular
functions.
Dividends linear in the surplus are classical in pension funding. In aggre-
gated portfolio models, such dividends were for a long time known to be
optimal, adopting a quadratic optimization criteria that punishes deviations
218 Surplus-linked life insurance
from targets of both payments and surplus. See Cairns (2000) for a state of the
art exposition of results in this respect. Furthermore, Steffensen (2006a) shows
that linear dividends minimize deviations on any sub-portfolio level, and
hence even on an individual basis. In addition, Steffensen (2004) shows that
in the case of participating life insurance, where dividends are constrained to
be positive, linear dividends are optimal, adopting a power utility optimiza-
tion criterion. This holds at least if dC(t) =0, t >0, which is the case if we
work with the systematic surplus and take the technical basis to be equal to
the market basis. Thus, there are several motivations for a special study of
linear dividends. See also Nielsen (2005, 2006) for further results on utility
optimal dividends in life insurance.
With dividends given in the form of Equation (6.17), Equations (6.16a–c)
turn into the following:
V
;
t
(t. x) =rV
;
(t. x) −l
;
(t) −q
;
(t) x−
¸
k:k=;
R
;k
(t. x) µ
;k
(t)
−V
;
x
(t. x)

rx+c
;
(t) −q
;
(t) x


1
2
u
;
(t. x)
2
x
2
V
;
xx
(t. x) . t ∈ |0. n] \. (6.18a)
V
;
(t−. x) =AB
;
(t) +AQ
;
(t) x+V
;

t. x
;

. t ∈ . (6.18b)
R
;k
(t. x) =l
;k
(t) +q
;k
(t) x+V
k

t. x
;k

−V
;
(t. x) . (6.18c)
x
;k
=x+c
;k
(t) −q
;k
(t) x.
x
;
=x+AC
;
(t) −AQ
;
(t) x.
We now guess that the solution to this differential equation can be written
in the following form:
V
;
(t. x) =1
;
(t) +g
;
(t) x. (6.19)
We can verify this solution and derive ordinary differential equations for 1
and g by plugging the solution, Equation (6.19), into Equations (6.18a–c). By
collecting terms with and without x and dividing the terms including x by x,
we arrive at two differential equations characterizing 1 and g, respectively.
Introducing
m
;
(t) =g
;
(t) c
;
(t) .
m
;k
(t) =g
k
(t) c
;k
(t) .
AE
;
(t) =g
;
(t) AC
;
(t) .
6.5 Dividends linear in surplus 219
the differential equation for 1 is as follows:
1
;
t
(t) =r1
;
(t) −l
;
(t) −m
;
(t) −
¸
k:k=;
R
E;k
(t) µ
;k
(t) . t ∈ |0. n] \.
(6.20)
1
;
(t−) =AB
;
(t) +AE
;
(t) +1
;
(t) . t ∈ .
R
E;k
(t) =l
;k
(t) +m
;k
(t) +1
k
(t) −1
;
(t) .
For a given function g, the structure of this differential equation is similar
to the structure of Equation (6.2). We can then conclude that 1 has the
following representation:
1
;
(t) =E
t.;
¸

n
t
e

s
t
r
d (B+E) (s)
¸
. (6.21)
dE (t) =dE
Z(t)
(t) +
¸
k∈
m
Z(t−)k
dN
k
(t) .
dE
;
(t) =m
;
(t) dt +AE
;
(t) . ; ∈ .
The function 1
;
(t) represented by Equation (6.21) contains two parts. The
first part is simply V
B;
(t), the market value of first order payments. The
second part,
E
t.;
¸

n
t
e

s
t
r
dE (s)
¸
.
represents the part of future contributions to the surplus that are expected
to be redistributed. This corresponds to the value of the payment process
C, where, however, the coefficients are weighted with the factor g and
then lead to coefficients of the artificial payment process E. If, in partic-
ular, we work with the systematic surplus and the market basis is used
as the technical basis, there are no future contributions and we have the
following:
V
;
(t. x) =V
B;
(t) +g
;
(t) x.
Introducing
r
G;
(t) =q
;
(t) +
¸
k:k=;
q
;k
(t) µ
;k
(t) .
220 Surplus-linked life insurance
the differential equation for g is as follows:
g
;
t
(t) =g
;
(t) r
G;
(t) −q
;
(t)

¸
k:k=;
R
G;k
(t)

1−q
;k
(t)

µ
;k
(t) . t ∈ |0. n] \. (6.22)
g
;
(t−) =

1−g
;
(t)

AQ
;
(t) +g
;
(t) . t ∈ .
R
G;k
(t) =q
;k
(t) ¡

1−q
;k
(t)

+g
k
(t) −g
;
(t) .
The structure of this differential equation is also similar to the structure
in Equation (6.2). However, we need to deal with two special circumstances
when writing down a conditional expected value representation of the solution.
Firstly, the intensity appears in Equation (6.22) with the factor 1−q
;k
(t). This
is, however, just a matter of taking the expectation under an appropriately
chosen measure. Secondly, the pasting condition specifies the lump sum
payment

1−g
;
(t)

AQ
;
(t). Thus, the function g itself appears in the lump
sum payment at deterministic points in time. We now conclude that g has the
following representation:
g
;
(t) =E
q
t.;
¸

n
t
e

s
t
r
g
dG(s)
¸
. (6.23)
dG(t) =dG
Z(t)
(t) +q
Z(t−)k
(t) ¡

1−q
Z(t−)k
(t)

dN
k
(t) .
dG
;
(t) =q
;
(t) dt +AG
;
(t) .
AG
;
(t) =

1−g
;
(t)

AQ
;
(t) .
where E
q
denotes the expectation with respect to a measure under which N
k
admits the intensity process

1−q
Z(t)k
(t)

µ
Z(t)k
(t).
Example 6.4 We consider the survival model with two states corresponding
to the policy holder being alive or dead. For simplicity, we restrict ourselves
to lump sum payments at deterministic time points at time n only, i.e. =]n].
Furthermore, we think of the realistic situation where no payments fall due
after occurrence of death. In this case, it is possible to work with an extensive
simplification of notation: N = N
1
, µ = µ
01
and, for all other quantities
and functions, the specification of state 0 is skipped, i.e. l = l
0
, l
1
= l
01
,
AB = AB
0
, 1 = 1
0
, g = g
0
, etc. We assume that all dividends are paid out
6.5 Dividends linear in surplus 221
only at the rate o(t) =q (t) X(t), i.e. o
1
(t) =AD(t) =0. Then functions 1
and g become
1 (t) =V
B
(t) +

n
t
e

s
t
r+µ
g (s) c (s) ds.
g (t) =

n
t
e

s
t
q+µ
q (s) ds.
If, in particular, we work with the systematic surplus and the market basis
and the technical basis coincide, we obtain
1 (t) =V
B
(t) .
g (t) =

n
t
e

s
t
q+µ
q (s) ds.
Remark 6.5 Continuing Remark 6.3, we consider the special case where
o(t. x) = q (t) x and AD(n. x) = AQ(n) x. Then the differential equa-
tions (6.20) and (6.22) reduce to
1
t
(t) = r1 (t) +r−g (t) c (t) .
1 (n−) = l (0) .
g
t
(t) = g (t) q (t) −q (t) .
g (n−) = AQ(n) .
These differential equations can be recognized from Sections 4.5.1 and 4.5.2.
In those sections these differential equations are given as special cases where
the investment is sufficiently prudent to secure that X(t) ≥0. The reason for
this was that the dividends were considered as something beneficial to the
policy holder. In the present chapter, we have made no such requirements.
The quantities called 8 in Sections 4.5.1 and 4.5.2 are put to zero since,
in the present chapter, we have chosen to present just linear and not affine
dividends.
The differential equations for 1 and g are presented in Section 4.5.1 for
terminal dividends under “prudent investment” with q =0 and AQ(n) =o.
The explicit solution is also presented there. The differential equations 1 and
g are presented in Section 4.5.2 for continuous dividends under “prudent
investment” with q (t) = o(t) and AQ(n) = 0. The explicit solution is also
presented there.
222 Surplus-linked life insurance
6.6 Bonus
In this section we consider the situation where dividends are not paid out
to the policy holder in cash but are converted into an additional stream of
bonus payments. One would typically constrain this conversion to happen on
a market basis, i.e. such that the additional insurance added and paid for by
the dividend payment is fair in itself. Then, in principle, there is no need
to pay special attention to the conversion. Nevertheless, we here present a
mathematical framework for handling bonus payments. In the following the
additional bonus payments are spoken of as additional first order payments,
although they may in practice not necessarily have the same status as the first
order payments initiated at time 0.
We now have first order payments and dividends from each insurance
contract which has been added in the past. We augment the functions by an
additional time argument which specifies the time of initiation. Thus, B(s. t)
and D(s. t) are the accumulated first order payments and dividend payments,
respectively, at time t stemming from contracts initiated during |0. s]. We
need to work with the dynamics of B(s. t) in both time dimensions. We
use the notation dB(s. t) for the dynamics of B(s. t) in the time of payment
dimension, i.e.
dB(s. t) =B(s. t) −B(s. t−) +l
t
(s. t) dt.
such that B
t
(s. t) = l
t
(s. t) wherever it exists, and we use the notation
B(ds. t) for the dynamics of B(s. t) in the time of initiation
dimension, i.e.
B(ds. t) =B(s. t) −B(s−. t) +l
s
(s. t) ds.
such that B
s
(s. t) =l
s
(s. t) wherever it exists. Thus, the differential dB(s. t)
denotes, conforming with previous sections, the change in accumulated
payments over (t −dt. t] for a given fixed time of initiation s. The dif-
ferential B(ds. t) denotes the change in accumulated payments at time t
stemming from initiation of payments over (s −ds. s]. The relations in this
paragraph indicate that the results in this section are derived for payment
processes with the same structure as the payment processes in the preceding
sections.
6.6 Bonus 223
We now denote by B(t) and D(t) the total accumulated first order pay-
ments and dividend payments, respectively, at time t, stemming from all
insurance contracts initiated in the past. We can then write the following:
B(t) =B(0−. t) +

t
0−
B(ds. t) .
D(t) =D(0−. t) +

t
0−
D(ds. t) .
Here B(0−. ·) and D(0−. ·) are the payment processes stemming from the
original contract, corresponding to the payment processes B and D in the
previous sections.
The circumstance that the total dividend payment paid at time t is used as
a single premium of an insurance contract initiated at time t can be written
as follows:
dD(t) =−B(dt. t) . (6.24)
from which we derive the dynamics of the total accumulated payments B+D:
dB(t) +dD(t) =dB(0−. t) +d

t
0−
B(ds. t) +dD(t)
=dB(0−. t) +B(dt. t) +

t−
0−
dB(ds. t) +dD(t)
=dB(0−. t) +

t−
0−
dB(ds. t) . (6.25)
This simply shows that the total payments that the policy holder pays and
receives are equal to the sum over all first order payments bought until
and excluding time t. This is how far we get without specifying further the
additional bonus payment. However, we have put some structure on the bonus
payment process which allows for substantial mathematical progress.
We now restrict ourselves to the situation where, at any point in time t, the
dividends are used to buy a fraction of the part of a given payment process A
due over (t. n]. The structure of the payment process A is assumed to coincide
with the structure of the original first order payment process B(0−. ·). Actu-
ally, Acould be equal to B(0−. ·). It could also consist of premiums or benefits
of B(0−. ·) exclusively. In Steffensen (2001), some calculations are made for
A = B(0−. ·). In Ramlau-Hansen (1991) and Norberg (1999), some calcula-
tions are made for A = B
+
(0−. ·), where the superscript + specifies that only
benefits are taken into account. The different possibilities of Aare related to the
notions of defined contributions and defined benefits in the sense that A could
224 Surplus-linked life insurance
be the part of B which is not defined but is instead adapted to the development of
the contract. We introduce C
B
(t) and C
A
(t) to denote the accumulated surplus
contributions based on the payment process Aand B(0−. ·), respectively.
Denoting by dY (s) the fraction of the payment process A bought at time s,
we can formalize the idea explained in the previous paragraph, for 0 -s -t,
as follows:
B(ds. t) =B(ds. s) +dY (s) (A(t) −A(s)) .
or, in differential form,
dB(ds. t) =dY (s) dA(t) . (6.26)
By the convention Y (0−) =0, the process Y represents the total proportion
of A bought until time t, and using Equation (6.26) we can then continue the
calculation in Equation (6.25) as follows:
dB(t) +dD(t) =dB(0−. t) +

t−
0−
dY (s) dA(t)
=dB(0−. t) +Y (t−) dA(t) .
We also need an extension of notation for the reserves. We let V
B∗Z(t)
(s. t)
and C(s. t) denote the technical reserve and the accumulated contributions
to the surplus, respectively, at time t for contracts initiated until time s. By
application of Equation (6.26), one can now verify the following relations:
V
B∗Z(t)
(ds. t) ≡E
¸

n
t
e

s
t
r
dB(ds. t)

(t)
¸
(6.27)
=dY (s) V
A∗Z(t)
(t) .
dC(ds. t) =dY (s) dC
A
(t) .
Conforming with the notation A and B, we let C denote the total contributions
stemming from all contracts initiated in the past. These contributions are as
follows:
C(t) =C(0−. t) +

t
0−
C(ds. t) .
dC(t) =dC(0−. t) +d

t
0−
C(ds. t)
=dC
B
(t) +C(dt. t) +

t−
0−
dC(ds. t)
=dC
B
(t) +C(dt. t) +Y (t−) dC
A
(t) .
6.6 Bonus 225
We now assume that the first order basis on which additional first order
payment processes are determined is used as the technical basis in the surplus.
Then we have the following two relations from the equivalence principle and
Equation (6.14), respectively:
V
B∗Z(t)
(dt. t) = −B(dt. t) . (6.28)
C(dt. t) =0. (6.29)
We can now, using Equation (6.27) with s = t, Equation (6.28) and Equa-
tion (6.24), relate the dynamics of Y and the dynamics of D as follows:
dY (t) =
V
B∗Z(t)
(dt. t)
V
A∗Z(t)
(t)
=−
B(dt. t)
V
A∗Z(t)
(t)
=
dD(t)
V
A∗Z(t)
(t)
. (6.30)
A particular dividend plan relating to the so-called contribution plan (see
Norberg (1999)), specifies that D=C. Plugging this plan into Equation (6.30)
leads to the following stochastic differential equation for Y:
dY (t) =
dC(t)
V
A∗Z(t)
(t)
=
dC
B
(t) +Y (t−) dC
A
(t)
V
A∗Z(t)
(t)
.
which is a generalized version of corresponding differential equations in
Norberg (1999) and Ramlau-Hansen (1991).
Instead, we study in Section 6.7 dividends partly similar to those in Equa-
tions (6.15). For this, we need the dynamics of the total surplus, still given by
dX(t) =rX(t) dt +u
Z(t)
(t. X(t) . Y (t)) X(t) dW (t) +d (C−D) (t) .
Here u
Z(t)
(t. X(t) . Y (t)) is the proportion of X invested in S
1
, now allowed
to depend also on Y. We take u
;
(t. x. x) to be a deterministic and sufficiently
regular function.
226 Surplus-linked life insurance
Remark 6.6 The arrangement of payments presented in this section is a
generalization of the arrangement that was presented in Chapter 2. Consider
again the endowment insurance from Remarks 6.1 and 6.2. Let the payment
process A be given by a pure endowment, i.e. dA(t) =dA
Z(t)
(t) and dA
0
(t) =
de
n
(t) . dA
1
(t) = 0. Then the differential equation for the total technical
reserve corresponding to the alive state is, skipping the superfluous state
specification, given by
d
dt
V
B∗
(t) =
d
dt

V
B∗
(0. t) +Y (t) V
A∗
(t)

d
dt
V
B∗
(0. t) +
d
dt
Y (t) V
A∗
(t) +Y (t)
d
dt
V
A∗
(t)
=r

V
B∗
(0. t) +r−µ

(t)

l
ad
−V
B∗
(0. t)

+o(t) +Y (t)

r

V
A∗
(t) −µ

(t)

−V
B∗
(0. t)

=r

V
B∗
(t) +r−µ

(t)

l
ad
−V
B∗
(t)

+o(t) .
Due to the equivalence principle, Equation (6.4), and Y (0) =0. V
B∗
(t) ful-
fils the initial condition V
B∗
(0) = 0. Hereafter, the differential equation for
V
B∗
(t) is recognized as the differential equation for the technical reserve
presented in Section 2.2.
The statewise surplus contribution rate is given by
c (t) = c
B
(t) +Y (t) c
A
(t)
=(r −r

) V

(0. t) +(µ

(t) −µ(t))

l
ad
−V
B∗
(0. t)

+Y (t)

(r −r

) V
A∗
(t) +(µ

(t) −µ(t))

−V
A∗
(t)

=(r −r

) V
B∗
(t) +(µ

(t) −µ(t))

l
ad
−V
B∗
(t)

.
which is exactly the contribution rate presented in Equation (2.16). The link
to the undistributed reserve presented in Section 2.2.3 now boils down to
noting that the undistributed reserve is defined as the mean portfolio surplus
per expected, as explained in Remark 6.2.
6.7 Surplus- and bonus-linked dividends
In this section we study, within the framework presented in Section 6.6, a par-
ticular dividend process, namely a dividend process where the dividend pay-
ments are linked to the surplus X and to the process Y defined in Section 6.6.
6.7 Surplus- and bonus-linked dividends 227
We formalize this by defining the processes o
;
(t), AD
;
(t) and o
;k
(t) as
functions of (X(t) . Y (t)):
o
;
(t) =o
;
(t. X(t) . Y (t)) .
o
;k
(t) =o
;k
(t. X(t) . Y (t)) . (6.31)
AD
;
(t) =AD
;
(t. X(t) . Y (t)) .
where we, by abuse of notation, use the same letters for the processes and the
functions.
The motivation for this study is the same as in Section 6.4. Firstly, we
suggest the formalization for valuation, management and supervision in par-
ticipating life insurance and pension funding. Secondly, we suggest it for
product design in life and pension insurance. However, the fact that we allow
dividends to depend also on Y requires comment.
The idea is to obtain a stabilizing effect, for example by letting the
dividends payments be relatively large when Y is relatively small, all other
things being equal. This idea has recently been practised by a major Danish
life insurance company in the following extreme sense. Dividends are set
in relation to the surplus X, and even negative dividends are allowed to be
paid out. However, one is not allowed to violate the first order payments
determined at time 0−, and for this reason one is only allowed to collect
negative dividends as long as Y is positive. A formalization of this practice
could, for example, be o
;
(t) = q
;
(t) X(t) I (Y (t) ≥0). The allowance for
collecting negative dividends and hereby lowering first order payments can
be spoken of as “conditional bonus.”
We refer again to Steffensen (2000, 2001) for the deterministic differential
equation that characterizes the reserve V under Equations (6.31). The state
process on which all payments depend is, in this section, (Z(t) . X(t) . Y (t)).
In Section 6.4, we argued that (Z(t) . X(t)) is Markovian. Informally,
(Z(t) . X(t) . Y (t)) is then seen to be Markovian since all coefficients of
the stochastic differential equation for Y depend on (Z(t) . X(t) . Y (t)) only;
see Equations (6.30), (6.5) and (6.31). From Steffensen (2000, 2001), we see
that the deterministic function
V
;
(t. x) =E
¸

n
t
e

s
t
r
dB(s)

Z(t) =;. X(t) =x. Y (t) =x
¸
228 Surplus-linked life insurance
is fully characterized by the following differential equation:
V
;
t
(t. x. x) =rV
;
(t. x. x) −l
;
(t) −xo
;
(t) −
¸
k:k=;
R
;k
(t. x. x) µ
;k
(t)
−V
;
x
(t. x. x)

rx+c
B;
(t) +xc
A;
(t) −o
;
(t. x. x)

−V
;
x
(t. x. x)

o
;
(t. x. x)
V
A∗;
(t)


1
2
u
;
(t. x. x)
2
x
2
V
;
xx
(t. x. x) . t ∈ |0. n] \. (6.32)
V
;
(t−. x. x) =AB
;
(t) +xAA
;
(t) +V
;

t. x
;
. x
;

. t ∈ .
R
;k
(t. x. x) =l
;k
(t) +xo
;k
(t) +V
k

t. x
;k
. x
;k

−V
;
(t. x. x) .
x
;k
=x+c
B;k
(t) +xc
A;k
(t) −o
;k
(t. x. x) .
x
;k
=x +
o
;k
(t. x. x)
V
A∗k
(t)
.
x
;
=x+AC
B;
(t) +xAC
A;
(t) −AD
;
(t. x. x) .
x
;
=x +
AD
;
(t. x. x)
V
A∗;
(t)
.
The differential equation (6.32) can be used to calculate a set of fair
functions, Equations, (6.15) leading to equivalence in the sense of
V
0
(0−. 0. 0) =0.
This corresponds to the equivalence relation in Equation (6.8). In general,
the differential equation (6.32) must be solved by numerical methods. In
special cases we can, however, reduce the partial differential equation to
a set of ordinary differential equations. Such a special case is studied in
Section 6.8.
Remark 6.7 Consider again the financial contract studied in Remarks 6.3
and 6.5. We now let dividends be used to buy additional elementary endow-
ment payments. Then Equation (6.32) reduces to
V
t
(t. x. x) =rV (t. x. x) +r (6.33)
−V
x
(t. x. x)

rx+c
B
(t) +xc
A
(t) −o(t. x. x)

−V
x
(t. x. x)

o(t. x. x)
V
A∗
(t)


1
2
u (t. x. x)
2
x
2
V
xx
(t. x. x) .
V (n−. x. x) =l (0) +x.
6.8 Dividends linear in surplus and bonus 229
We now translate this into a differential equation for another function
¯
V,
where the reserve is formalized as a function of time, surplus and the total
technical reserve. Thus, we put V (t. X(t) . Y (t)) =
¯
V

t. X(t) . V
B∗
(t)

. We
note first that
c
B
(t) +xc
A
(t) = (r −r

) V
B∗
(0. t) +(r −r

) xV
A∗
(t)
= (r −r

)
¨
V
B∗
(t. x) .
where
¨
V
B∗
(t. x) =V
B∗
(0. t) +xV
A∗
(t) .
i.e. V
B∗
(t) =
¨
V
B∗
(t. x( t)). Furthermore, using the following:
¨
V
B∗
t
(t. x) =V
B∗
t
(0. t) +xV
A∗
t
(t)
=r

V
B∗
(0. t) +r+xr

V
A∗
(t)
=r

¨
V
B∗
(t. x) +r.
we obtain the translation of relevant partial derivatives:
V
x
=
¯
V
v
¨
V
B∗
x
(t. x) =
¯
V
v
V
A∗
(t) .
V
t
=
¯
V
t
+
¯
V
v
¨
V
B∗
t
(t. x) =
¯
V
t
+
¯
V
v

r

¨
V
B∗
(t. x) +r

.
Plugging these relations into Equation (6.33), replacing
¨
V
B∗
(t. x) by v and
rearranging slightly, we arrive at
¯
V
t
(t. x. v) =r
¯
V (t. x. v) +r−
¯
V
x
(t. x. v) (rx+(r −r

) v −o(t. x. x))

¯
V
v
(r

v +r+o(t. x. x)) −
1
2
u (t. x. x)
2
x
2
¯
V
xx
(t. x. v) .
¯
V (n−. x. x) =l (0) +x.
It is obvious that this is the partial differential equation given in Section 4.5.3
for the characterization of the reserve.
6.8 Dividends linear in surplus and bonus
In this section, we specify further the dividend formalization suggested
in Equations (6.31). We derive a semi-explicit solution to the differential
230 Surplus-linked life insurance
equation (6.32) in the case where the dividend payments are linear functions
of the surplus X and the process Y, i.e.
o
;
(t) =q
;
(t) X(t) +µ
;
(t) Y (t) .
o
;k
(t) =q
;k
(t) X(t) +µ
;k
(t) Y (t) . (6.34)
AD
;
(t) =AQ
;
(t) X(t) +Ao
;
(t) Y (t) .
where q
;
(t), q
;k
(t), AQ
;
(t), µ
;
(t), µ
;k
(t) and Ao
;
(t) are deterministic and
sufficiently regular functions. As in Section 6.5, dividends linear in surplus
can be motivated by practice and by certain optimization problems. Linking
dividends to the process Y is encouraged in Section 6.7, and below we show
that a linear link is one construction which leads to a semi-explicit solution.
With dividends given in the form of Equation (6.34), Equation (6.32) turns
into the following differential equation:
V
;
t
(t. x. x) =rV
;
(t. x) −l
;
(t) −xo
;
(t) −
¸
k:k=;
R
;k
(t. x. x) µ
;k
(t)
−V
;
x
(t. x)

rx+c
B;
(t) +xc
A;
(t) −q
;
(t) x−µ
;
(t) x

−V
;
x
(t. x. x)

q
;
(t) x+µ
;
(t) x
V
A∗;
(t)


1
2
u
;
(t. x. x) x
2
V
;
xx
(t. x. x) . t ∈ |0. n] \. (6.35)
V
;
(t−. x. x) =AB
;
(t) +xAA
;
(t) +V
;

t. x
;
. x
;

. t ∈ .
R
;k
(t. x. x) =l
;k
(t) +xo
;k
(t) +V
k

t. x
;k
. x
;k

−V
;
(t. x. x) .
x
;k
=x+c
B;k
(t) +xc
A;k
(t) −q
;k
(t) x−µ
;k
(t) x.
x
;k
=x +
q
;k
(t) x+µ
;k
(t) x
V
A∗k
(t)
.
x
;
=x+AC
B;
(t) +xAC
A;
(t) −AQ
;
(t) x−Ao
;
(t) x.
x
;
=x +
AQ
;
(t) x+Ao
;
(t) x
V
A∗;
(t)
.
We now guess that the solution to this differential equation can be written
in the following form:
V
;
(t. x. x) =1
;
(t) +g
;
(t) x+h
;
(t) x. (6.36)
6.8 Dividends linear in surplus and bonus 231
We can verify this solution and derive ordinary differential equations for 1,
g and h by plugging Equation (6.36) into the differential equation (6.35).
By collecting terms with x, with x and without x and x, and dividing terms
including x and x by x and x, respectively, we arrive at three differential
equations characterizing 1, g and h, respectively.
The differential equation for 1 coincides with Equation (6.20) with c
;
(t)
and c
;k
(t) replaced by c
B;
(t) and c
B;k
(t). With this replacement, the represen-
tation in Equation (6.21) also holds true with E replaced by an appropriately
defined artificial payment process E
B
. The latter part of this representation
again represents the part of future contributions to the surplus which belongs
to the policy holder, now after conversion into bonus payments.
The differential equation for g is exactly the same as Equation (6.22)
with q
;
(t) and q
;k
(t) in the payment process replaced by the fractions
q
;
(t) h
;
(t) ¡V
A∗;
(t) and q
;k
(t) h
k
(t) ¡V
A∗k
(t), respectively. In the artificial
interest rate r
G;
(t), q
;
(t) and q
;k
(t) should not be replaced. With this replace-
ment the representation in Equation (6.23) also holds true, i.e. for
dG(t) =dG
Z(t)
(t) +
q
Z(t−)k
(t) h
k
(t) ¡V
A∗k
(t)
1−q
Z(t−)k
(t)
dN
k
(t) .
dG
;
(t) =q
;
(t) h
;
(t) ¡V
A∗;
(t) dt +AG
;
(t) .
AG
;
(t) =

1−g
;
(t)

AQ
;
(t) .
Again, the function g specifies the part of the surplus that belongs to the
policy holders. However, now the elements in this function are weighted with
the fraction h
Z(t)
(t) ¡V
A∗Z(t)
(t).
Introducing the following:
r
;
(t) =m
A;
(t) −g
;
(t) µ
;
(t) .
r
;k
(t) =m
A;k
(t) −g
k
(t) µ
;k
(t) .
AH
;
(t) =AE
A;
(t) −g
;
(t) Ao
;
(t) +h
;
(t)
Ao
;
(t)
V
A∗;
(t)
.
r
H;
(t) =r −
µ
;
(t)
V
A∗;
(t)
+
¸
k:k=;
µ
;k
(t)
V
A∗k
(t)
µ
;k
(t) .
232 Surplus-linked life insurance
the differential equation for h is given by
h
;
t
(t) =r
H;
(t) h
;
(t) −o
;
(t) −r
;
(t)

¸
k:k=;
R
H;k
(t)
V
A∗k
(t) +µ
;k
(t)
V
A∗k
(t)
µ
;k
(t) . t ∈ |0. n] \. (6.37)
h
;
(t−) =AA
;
(t) +AH
;
(t) +h
;
(t) . t ∈ .
R
H;k
(t) =

o
;k
(t) +r
;k
(t)

V
A∗k
(t)
V
A∗k
(t) +µ
;k
(t)
+h
k
(t) −h
;
(t) .
The structure in this differential equation is similar to the structure in Equa-
tion (6.22). However, as in Equations (6.22) and (6.23) we need to deal with
two special circumstances when writing down a conditional expected value
representation of the solution. Firstly, the intensity appears in Equation (6.37)
with the factor

V
A∗k
(t) +µ
;k
(t)

¡V
A∗k
(t), but this is just a matter of taking
the expectation under an appropriately chosen measure. Secondly, the pasting
condition specifies a lump sum payment that includes h
;
(t) Ao
;
(t) ¡V
A∗;
(t).
Thus, the function h itself appears in the lump sum payment at deterministic
points in time. We now conclude that h has the following representation:
h
;
(t) =E
µ
¸

n
t
e

s
t
r
´
H
d(A+H) (s)
¸
.
dH(t) =dH
Z(t)
(t) +r
Z(t−)k
(t)
V
A∗k
(t)
V
A∗k
(t) +µ
;k
(t)
dN
k
(t) .
dH
;
(t) =r
;
(t) dt +AH
;
(t) .
where E
µ
denotes the expectation with respect to a measure under which
N
k
admits the intensity process

V
A∗k
(t) +µ
Z(t)k
(t)

¡V
A∗k
(t) µ
Z(t)k
(t). The
function h specifies the part of Y that belongs to the policy holder accord-
ing to the dividend and the bonus plans. Note that since g appears in the
differential equation for h and vice versa, these functions must be calculated
simultaneously.
In the special case where dividends are not linked to the process Y, i.e.
µ
;
= µ
;k
= Ao
;
= 0, we have that r
H;
(t) = r and H is similar to E except
that B and C
B
are replaced by A and C
A
, respectively. We conclude that 1
and h in this case differ by their underlying payment process only.
Example 6.8 We consider again the survival model introduced in Exam-
ple 6.4. We consider the same contract, now specified with bonus payments
6.8 Dividends linear in surplus and bonus 233
proportional to the payment process A. We assume that µ
;

;k
=Ao
;
=0,
leading to the following specialization of the above results:
1 (t) =V
B
(t) +

n
t
e

s
t
r+µ
g (s) c
B
(s) ds.
g (t) =

n
t
e

s
t
q+µ
h(s)
V
A∗
(s)
q (s) ds.
h(t) =V
A
(t) +

n
t
e

s
t
r+µ
g (s) c
A
(s) ds.
If, in particular, we work with the systematic surplus, and the market basis
and the technical basis coincide, we obtain the following:
1 (t) =V
B
(t) .
g (t) =

n
t
e

s
t
q+µ
h(s)
V
A∗
(s)
q (s) ds.
h(t) =V
A
(t) .
Remark 6.9 Continuing Remark 6.7, we consider the special case where
o(t. x. x) =q (t) x and AD(n. x) =0. Recall that the continuous payment rate
q (t) in the differential equation for g, Equation (6.22), should be replaced
by q (t) h(t) ¡V
A∗
(t). Then we obtain the following differential equations for
1, g and h from Equations (6.20), (6.22) and (6.37):
1
t
(t) =r1 (t) +r−g (t) c
B
(t) .
1 (n−) =l (0) .
g
t
(t) =g (t) q (t) −q (t) h(t) ¡V
A∗
(t) .
g (n−) =0.
h
t
(t) =rh(t) −g (t) c
A
(t) .
h(n−) =1.
We now compare these differential equations with the differential equations
for the coefficient functions in Section 4.5.3. For this purpose, we write
¯
V
from Remark 6.7 as an affine function of the surplus and the total technical
234 Surplus-linked life insurance
reserve. Using V (t. X(t) . Y (t)) =
¯
V

t. X(t) . V
B∗
(t)

, we then obtain the
following relation between the coefficient functions in the two expositions:
1 (t) +g (t) X(t) +h(t) Y (t) =
¯
1 (t) +¯g (t) X(t) +
¯
h(t) V
B∗
(t. Y (t))
=
¯
1 (t) +
¯
h(t) V
B∗
(0. t)
+¯g (t) X(t) +
¯
h(t) V
A∗
(t) Y (t) .
For X(t) =x and Y (t) =x, this yields
1 (t) +g (t) x+h(t) x =
¯
1 (t) +
¯
h(t) V
B∗
(0. t) +g (t) x+
¯
h(t) V
A∗
(t) x.
Realizing that h(t) =
¯
h(t) V
A∗
(t), g (t) =¯g (t) and 1 (t) =
¯
1 (t) +
¯
h(t) V
B∗
(0. t), we can now find differential equations characterizing 1, g and h.
Firstly, we find the equation for
¯
h:
¯
h
t
(t) =
V
A∗
(t) h
t
(t) −V
A∗
t
(t) h(t)
(V
A∗
(t))
2
=
rh(t) −g (t) c
A
(t) −r

h(t)
V
A∗
(t)
= (r −r

)
¯
h(t) −g (t) (r −r

) .
¯
h(n−) =1.
Secondly, we find the differential equation for ¯g:
¯g
t
(t) =¯g (t) q (t) −q (t)
¯
h(t) .
¯g (n−) =0.
Thirdly, we find the differential equation for
¯
1:
¯
1
t
(t) =1
t
(t) −
¯
h
t
(t) V
B∗
(0. t) −
¯
h(t) V
B∗
t
(0. t)
=r
¯
1 (t) +r−
¯
h(t) r.
¯
1 (n−) =1 (n−) −
¯
h(n−) l (0) =0.
These differential equations for
¯
1, ¯g and
¯
h can now be recalled from the
special case of “prudent investment” in Section 4.5.3, there taking 8 =0 in
accordance with the linear dividends in this section.
7
Interest rate derivatives in insurance
7.1 Introduction
This chapter gives an introduction to interest rate derivatives and their use in
risk-management for life insurance companies. The first part of the chapter
recalls the definitions of swap rates, swaps, swaptions and related products
within a setting similar to the one studied in the previous chapters. Then
we describe some pricing methods that have been proposed in the literature.
There are a vast number of instruments available in the financial markets,
and there exist many different models for the pricing of these instruments;
see, for example, Brigo and Mercurio (2001), Musiela and Rutkowski (1997)
and Rebonato (2002). Our treatment of this area is rather minimal, and our
aim is simply to provide a brief introduction to certain developments that
seem useful in an analysis of the risk faced by life insurance companies.
The reader is therefore referred to the abovementioned references for a more
detailed and systematic treatment of the basic theory.
We end the chapter by giving possible applications of these instruments in
the area of risk management for a life insurance company facing insurance
liabilities that cannot be hedged via bonds in the market due to the very long
time horizon associated with the liabilities. Typically, insurance companies
are faced by insurance liabilities that extend up to sixty years into the future,
whereas the financial markets typically do not offer bonds that extend more
than thirty years into the future. In addition, the market for interest rate
derivatives (for example, interest rate swaps) has become much more liquid
and can therefore be used effectively by insurance companies. Thus, it is
relevant to investigate how a combination of investments in bonds of a
relatively short maturity and advanced (long term) interest rate derivatives
can be used to reduce the risk associated with the liabilities.
235
236 Interest rate derivatives in insurance
7.2 Swaps and beyond in continuous time
What is a swap rate? What is the relationship between swap rates, forward
rates and short rates? What are swaps, swaptions and CMS options?
Below we present a discussion of these and other concepts. Our presenta-
tion differs slightly from most other presentations of this topic in that we set
out by introducing the basic concepts in a continuous-time framework. This
has the advantage that it simplifies notation and is more closely related to
the previous chapters. We go on to discuss the corresponding quantities in a
discrete-time setting, which corresponds to the approach taken in practice.
As in Chapter 3, we denote by r(t) the continuously compounding short
rate at time t, and we let F be some information flow (filtration) related to
the bond market. Note that Q is the market measure, i.e. the probability mea-
sure used for the pricing of bonds and interest rate derivatives. In particular,
the price at time t of a zero coupon bond expiring at time T is given by
P(t. T) =E
Q
¸
e

T
t
r(s) ds

(t)
¸
=e

T
t
1(t.s) ds
.
where (1(t. s))
t≤s≤T
is the forward rate curve at time t.
Consider a general bond with payment process C of finite variation, which
is adapted to the information flow F. This means that payments during a
small time interval (t. t +dt] are given by dC(t).
Bonds with continuously paid coupons
As a main example, we study a bond with continuously paid coupons with
rate c(t) and a possible lump sum payment at time T. This leads to the
following payment process:
dC(t) =c(t) dt +AC(T) de(t. T). (7.1)
where e(t. T) = 1
]t≥T]
and where AC(T) is the possible payment at time T.
(Recall the notation introduced in Chapter 2.) The present value of all (past
and future) payments from C at time 0 ≤t ≤T is given by
PV(t. C) =

T
0
e

t
t
r(s) ds
dC(t) (7.2)
=

T
0
e

t
t
r(s) ds
c(t) dt +e

T
t
r(s) ds
AC(T).
and the market value (or price) at time t of future payments is given by
r(t. C) =E
Q
¸

T
t
e

t
t
r(s) ds
dC(t)

(t)
¸
.
7.2 Swaps and beyond in continuous time 237
One important example of a payment process of the form in Equation (7.1)
is where c(t) is identical to the short rate r(t) and AC(T) = 1. This can be
viewed as a continuous-time bullet bond with a variable interest rate similar
to the one studied in discrete time in Chapter 3. In this case, the present value
at time t of the future payments from C is given by

T
t
e

t
t
r(s) ds
r(t) dt +e

T
t
r(s) ds
. (7.3)
Since

T
t
e

t
t
r(s) ds
r(t) dt =−e

t
0
r(s) ds

T
t
de

t
0
r(s) ds
=1−e

T
t
r(s) ds
.
we see that Equation (7.3) is constant and equal to unity. Thus, the payment
process with (c. AC(T)) = (r. 1) has the special property that the present
value at any time t -T of future payments is constant and equal to unity.
Another important example is where c(t) is constant and equal to some
value « and where AC(T) = 1. In this case, the present value of future
payments becomes
«

T
t
e

t
t
r(s) ds
dt +e

T
t
r(s) ds
. (7.4)
The market value at time t of the future payments can be expressed in terms
of the zero coupon bond prices P(t. t):
«

T
t
P(t. t) dt +P(t. T); (7.5)
see Chapters 3 and 5 for similar calculations.
Swaps and swap rates
One can now look for the value of « which ensures that the prices of the two
payment streams coincide at a fixed time t, i.e. the value of « which ensures
that Equation (7.5) is identical to unity. It follows by simple calculations that
this is obtained for the following:
«(t. T) =
1−P(t. T)

T
t
P(t. t) dt
. (7.6)
We refer to «(t. T) as the (T −t)-year swap rate at time t (under con-
tinuous payments). For this particular value of «, market prices at time t
of future payments from the two payment streams coincide. This implies
that two parties can exchange (swap) these payment streams at time t with-
out making additional payments. The contract which consists in paying the
coupons c =r and receiving the coupons c =« is called a receiver swap. The
238 Interest rate derivatives in insurance
contract in which the owner receives c =r and pays c =« is called a payer
swap.
Relationships between swap rates and forward rates
We can obtain an alternative expression for the swap rate by rewriting the
term 1−P(t. T) in Equation (7.6) as follows:
1−P(t. T) =1−e

T
t
1(t.t) dt
=

T
t
1(t. t) e

t
t
1(t.s) ds
dt =

T
t
1(t. t)P(t. t) dt.
By inserting this into Equation (7.6), we obtain the following expression:
«(t. T) =

T
t
1(t. t)P(t. t) dt

T
t
P(t. t) dt
. (7.7)
which shows that the swap rate at time t is a weighted average of the forward
rates (1(t. t))
t≤t≤T
, weighted by the corresponding zero coupon bond prices.
We can now define the swap rate for each 0 ≤t ≤T and study the connection
between the forward rate curve and the swap curve («(t. t))
t≤t≤T
more closely.
Firstly, we use Equation (7.6) to derive a simple recursion for the zero
coupon bond prices. If we approximate the integral appearing in the denomi-
nator by the following:

T−h
t
P(t. t) dt +P(t. T)h
for small h, we see that
P(t. T) ≈
1−«(t. T)

T−h
t
P(t. t) dt
1+«(t. T)h
.
In this way, zero coupon bond prices may be determined recursively from a
given swap curve.
We know from Equation (7.7) that the swap rates are weighted averages
of the zero coupon bond prices. It can be seen that the zero coupon bond
prices can be expressed alternatively in terms of the swap curve via
P(t. T) =1−«(t. T)

T
t
e

T
t
«(t.s) ds
dt.
To verify this, insert this expression into Equation (7.6) and change the order
of integration in the denominator.
7.2 Swaps and beyond in continuous time 239
Caps and floors
As described above, the owner of a receiver swap receives the payments
dC(t) =(«−r(t)) dt, whereas the owner of the payer swap receives dC(t) =
(r(t)−«) dt. We now define two new contracts, known respectively as a cap
and a floor.
The owner of a cap receives payments of the form of Equation (7.1),
with c(t) =(r(t) −«)
+
and AC(T) =0; « is called the strike. The payments
from this contract are similar to the ones from the payer swap if the short
rate exceeds the strike, «, and zero otherwise. Let us modify the situation
slightly by considering the case where the payments start at some fixed time
T
0
- T. At time t - T
0
, the present value of the payments from the cap is
then given by

T
T
0
e

t
t
r(s) ds
(r(t) −«)
+
dt. (7.8)
A floor is similar, with c(t) = («−r(t))
+
and AC(T) = 0. In this case, the
payments are similar to those from a receiver swap if the short rate is smaller
than the strike «.
Swaptions
Consider as above the payments from a payer swap which starts at time T
0
and
stops at time T. The present value at time T
0
of these payments is given by

T
T
0
(r(t) −«) e

t
T
0
r(s) ds
dt.
which can be rewritten using calculations similar to the ones used above. In
particular, this shows that the price at time T
0
of the payer swap is given by
1−P(T
0
. T) −«

T
T
0
P(T
0
. t) dt.
A payer swap option (or payer swaption) is an option to enter a payer swap
at time T
0
with strike «. Thus, the payer swaption gives the holder the right,
but not the obligation, to enter the payer swap at time T
0
. The price of the
payer swaption at time t -T
0
can be calculated as follows:
E
Q
¸
e

T
0
t
r(s) ds

1−P(T
0
. T) −«

T
T
0
P(T
0
. t) dt

+
(t)

. (7.9)
At time t -T
0
, we would say that the maturity of the swaption is T
0
−t, the
tenor is T −T
0
and the strike is «. Similarly, a receiver swaption is an option
to enter a receiver swap.
240 Interest rate derivatives in insurance
Constant maturity swap (CMS) and CMS options
The swaption is really an option exercised at the fixed time T
0
where the
holder can enter the swap. At this time, the crucial quantity determining
whether or not there is a payoff from the option is the swap rate, «(T
0
. T).
A constant maturity swap (CMS) differs from the swap in that the payment
process involves the current swap rate with a fixed time T
0
to maturity. More
precisely, a CMS specifies payments dC(t) = («(t. T
0
+t) −«) dt during
|0. T
1
]. The present value at time 0 of these payments is given by

T
1
0
e

t
0
r(s) ds
(«(t. T
0
+t) −«) dt. (7.10)
The parameters are the number of years of payments T
1
, maturity of the swap
rates T
0
and the strike «.
A CMS cap is now obtained from Equation (7.10) by replacing the payment
intensity by («(t. T
0
+t) −«)
+
; a CMS floor is the case where the payment
intensity is («−«(t. T
0
+t))
+
. Thus, a CMS cap is essentially an integral
over call options on the T
0
-year swap rates at any time 0 ≤ t ≤ T. The
cap protects its owner against an increase in the T
0
-year swap rate during the
interval |0. T]. Similarly, a CMS floor involves putting options on the T
0
-year
swap rates and protecting its owner against a decrease in this swap rate.
7.3 Pricing of interest rate derivatives
7.3.1 Change of numeraire and forward measures
In the previous chapters we have been working under a martingale measure Q,
which had the special property that price processes discounted by the savings
account S
0
(t) =e

t
0
r(t) dt
were martingales under Q. In this case, we referred
to the savings account S
0
as the numeraire. The measure Q played a key role
in the calculation of arbitrage-free prices of derivatives and market values
of life insurance contracts, which were defined as expected present values
under Q. For the pricing of the various interest rate derivatives introduced
above, it is sometimes convenient to work with an alternative numeraire. If,
for example, we start by considering the present value of the payments from
the continuous-time version of the cap, we see from Equation (7.8) that the
integrand is given by
e

t
t
r(s) ds
(r(t) −«)
+
.
7.3 Pricing of interest rate derivatives 241
Thus, if we want to calculate the arbitrage-free price of the cap, we need to
determine
E
Q
¸
e

t
t
r(s) ds
(r(t) −«)
+

(t)
¸
. (7.11)
where we take t - t. This is not as straightforward as the standard Black–
Scholes formula for a European call option on a stock, since it involves the
joint distribution of

t
t
r(s) ds and r(t). One way of deriving a more simple
expression for the price, Equation (7.11), is to introduce another measure Q
t
,
which has the special property that all price processes discounted by the zero
coupon bond price process P(t. t) are martingales. Formally, this measure, the
forward measure, is defined via its Radon–Nikodym derivative as follows:
dQ
t
dQ
=
e

t
0
r(s) ds
P(0. t)
. (7.12)
Equation (7.11) can now be calculated by using the abstract Bayes rule. If,
for a moment, we use t = 0 and the notation g(r(t)) = (r(t) −«)
+
, we can
rewrite Equation (7.11) as follows:
E
Q
¸
e

t
0
r(s) ds
g(r(t))
¸
=P(0. t) E
Q
¸
e

t
0
r(s) ds
P(0. t)
g(r(t))

=P(0. t) E
Q
t
|g(r(t))] .
where we have simply divided by P(0. t) and multiplied by the same factor
in the first equality; in the second equality, we have used Equation (7.12), the
definition of the measure Q
t
. Here, it is important to note that the last expected
value is calculated under the new measure Q
t
, which typically differs from
the original martingale measure Q. This formula can be generalized to values
of t not necessarily equal to zero, and this leads to P(t. t) E
Q
t
| g(r(t)) (t)],
such that the price of the cap becomes

T
T
0
P(t. t) E
Q
t
¸
(r(t) −«)
+

(t)
¸
dt.
Again, it is important to note that the expected values are to be calculated
under the measures Q
t
for t ∈ |T
0
. T]. The cap price formula is even more
complicated, since it involves a new measure for each t. A similar formula can
be put up for the interest rate floor, the only difference being that (r(t) −«)
+
should be replaced by («−r(t))
+
.
It is not immediately clear that this is a useful method, since the distribution
of r(t) under Q
t
could be rather complicated. However, below we present
some well known examples for which the forward measure is not more
242 Interest rate derivatives in insurance
complicated than the risk neutral measure Q. First, recall that the forward rate
at time 0 for time t is defined as 1(0. t) = −(o¡ot) logP(0. t). Using this
defining relation, it can be shown that
E
Q
t
|r(t)] =E
Q
¸
e

t
0
r(s) ds
P(0. t)
r(t)

=−
o
ot
P(0. t)
P(0. t)
=−
o
ot
logP(0. t) =1(0. t).
Thus, the forward measure can be used to find the forward rates as expected
values of the short rates. This result can also be generalized such that 1(t. t) =
E
Q
t
|r(t)(t)].
The goal is now to describe in more detail the measures Q
t
, t ∈ |T
0
. T],
under some specific interest rate models, and to put up a closed formula
for the cap and floor prices obtained above. In general, a measure Q
t
is
completely determined by its density process:
Z
t
(t) =E
Q
¸
e

t
0
r(s) ds
P(0. t)

(t)

=
P(t. t)
P(0. t)
e

t
0
r(s) ds
. (7.13)
The Vasiˇ cek model
Consider the Vasiˇ cek model, where the dynamics of r under the market
martingale measure Q are given by
dr(t) =(l −or(t)) dt +u dW(t).
where W is a standard Brownian motion under Q. The price P(t. T) of a zero
coupon bond maturing at time T is given by
P(t. T) =e
A(t.T)−B(t.T)r(t)
. (7.14)
where A(t. T) and B(t. T) are determined in Chapter 3. It follows, by a
straight-forward application of Itô’s formula, that the dynamics of P(t. T)
under Q is given by
dP(t. T) =r(t)P(t. T) dt +v(t. T)P(t. T) dW(t). (7.15)
where v(t. T) = −uB(t. T). In this model, we have a very simple structure
for the forward rates (see Chapter 3):
1(t. T) =r(t) e
−o(T−t)
+

l
o

u
2
2o
2

1−e
−o(T−t)

+e
−o(T−t)
2u
2
B(t. T)
4o
.
Thus, the entire forward rate curve is determined from the short rate r. Using
Equation (7.7), we can now calculate the swap rate curve for a given forward
7.3 Pricing of interest rate derivatives 243
0 5 10 15 20 25 30
0.00
0.02
0.04
0.06
0.08
maturity
r
a
t
e
s
Figure 7.1. Forward rate curve (dashed line) and the corresponding swap rate
curve (solid line). Calculations for the Vasiˇ cek model with r(0) =0.03, u =0.03,
o =0.36 and l¡o =0.06.
rate curve. In Figure 7.1, the swap rate curve and the forward rate curve can
be found for the Vasiˇ cek model with the same parameters as in Chapter 3.
We see from Figure 7.1, that the swap curve in our situation is below the
forward rate curve. This can be explained by the fact that the forward rate
curve is increasing in this example.
In order to determine the behavior of r under the forward measure Q
T
, we
need to examine the density process Z
T
(t) and derive its dynamics. Using the
above expression for the zero coupon bond price, we see that
dZ
T
(t) =v(t. T)Z
T
(t) dW(t).
such that
dQ
T
dQ
=exp

T
0
v(t. T) dW(t) −
1
2

T
0
(v(t. T))
2
dt

.
Girsanov’s theorem now yields that the dynamics of r under Q
T
is given by
dr(t) =(l +uv(t. T) −or(t)) dt +u dW
T
(t). (7.16)
where W
T
is a standard Brownian motion under Q
T
. Thus, the equation with
the dynamics under Q
T
is similar to the one for the dynamics under our usual
martingale measure Q. One difference is that the drift term in Equation (7.16)
now involves the function v(t. T). This new type of dynamics should now be
used when calculating expected values like the ones appearing in the prices
for caps and floors.
244 Interest rate derivatives in insurance
Using Itô’s formula, we can verify the following:
r(T) =r(t) e
−o(T−t)
+

T
t
e
−o(T−s)
(l +uv(s. T)) ds
+

T
t
e
−o(T−s)
u dW
T
(s). (7.17)
which shows that the conditional distribution of r(t) given r(t) under Q
t
is
normal with mean equal to the forward rate 1(t. t) and variance given by
u
2

t
t
e
−2o(t−s)
ds =
u
2
2o

1−e
−2o(t−t)

=l
r
(t. t).
Thus, we have a very simple structure for the conditional distribution of the
short rate. However, even though the short rate is normally distributed, we
are not able to arrive at closed form expressions for the caps, floors and
swaptions introduced above. Instead, we can put up simple expressions for
the prices, which may be evaluated numerically.
For the cap, the price given in Equation (7.11) appearing under the integral
sign in the cap price can be written as follows:
P(t. t)E
Q
t
| g(r(t)) (t)] =P(t. t)
1

2rl
r
(t. t)
×


g(u) exp


(u−1(t. t))
2
2l
r
(t. t)

du. (7.18)
with g(u) =(u−«)
+
. This integral has to be evaluated numerically.
Similarly, the price given in Equation (7.9) of the payer swaption can be
determined as follows:
P(t. T
0
) E
Q
T
0
¸

1−P(T
0
. T) −«

T
T
0
P(T
0
. t) dt

+
(t)

.
where one can insert Equation (7.14) to obtain an expression similar to
Equation (7.18) and evaluate the integral numerically.
Finally, the price of the CMS floor can also be calculated numerically via
the following expression:

T
1
0
P(0. t) E
Q
t
¸
(«−«(t. T
0
+t))
+
¸
dt.
where «(t. T
0
+t) is given by either Equations (7.6) or (7.7).
7.3 Pricing of interest rate derivatives 245
Bond options in the Vasiˇ cek model
We finally provide a formula for call options on zero coupon bonds. We want
to calculate the price at time t of a call option maturing at time T
1
on a zero
coupon bond with maturity T
2
. This contract has the following price process:
H
call
(t) =E
Q
¸
e

T
1
t
r(s) ds
(P(T
1
. T
2
) −K)
+

(t)
¸
=E
Q
¸
e

T
1
t
r(s) ds
P(T
1
. T
2
)1
]P(T
1
.T
2
)≥K]

(t)
¸
−E
Q
¸
e

T
1
t
r(s) ds
K1
]P(T
1
.T
2
)≥K]

(t)
¸
.
Using the forward measure techniques introduced above, this can be rewritten
as follows:
H
call
(t) =P(t. T
2
)Q
T
2
( P(T
1
. T
2
) ≥K (t))
−KP(t. T
1
) Q
T
1
( P(T
1
. T
2
) ≥K (t)) . (7.19)
where we are using the forward measures Q
T
1
and Q
T
2
. Under Q
T
1
, we have
that P(t. T
2
)¡P(t. T
1
) is a Q
T
1
-martingale, and similarly P(t. T
1
)¡P(t. T
2
) is a
martingale under the measures Q
T
2
. Moreover, it follows from the dynamics
for the option price process, Equation (7.15), that
P(t. T
1
)
P(t. T
2
)
=
P(t. T
1
)
P(t. T
2
)
×exp

t
t
(v(s. T
1
) −v(s. T
2
)) dW(s)

1
2

t
t
(v(s. T
1
)
2
−v(s. T
2
)
2
) ds

. (7.20)
which involves W, a standard Brownian motion under Q. Alternatively, we
may use the relations between the dynamics under Q and under the forward
measures Q
T
i
, which imply that dW(t) =dW
T
i
(t) +v(t. T
i
) dt, where W
T
i
are
standard Brownian motions under Q
T
i
. This can be used to show that
P(t. T
1
)
P(t. T
2
)
=
P(t. T
1
)
P(t. T
2
)
exp

t
t
(v(s. T
1
) −v(s. T
2
)) dW
T
2
(s)

1
2

t
t
(v(s. T
1
) −v(s. T
2
))
2
ds

under Q
T
2
and
P(t. T
2
)
P(t. T
1
)
=
P(t. T
2
)
P(t. T
1
)
exp

t
t
(v(s. T
2
) −v(s. T
1
)) dW
T
1
(s)

1
2

t
t
(v(s. T
2
) −v(s. T
1
))
2
ds

246 Interest rate derivatives in insurance
under Q
T
1
. These two ways of writing the ratios between P(t. T
2
) and P(t. T
1
)
differ from Equation (7.20) in that they involve standard Brownian motions
under the new measures Q
T
1
and Q
T
2
, respectively. The two probabilities
appearing in Equation (7.19) can now be calculated by using these two
formulas. In fact, it follows that
Q
T
1
( P(T
1
. T
2
) ≥K (t)) =4(z
1
(t. T
1
. T
2
. K)) .
where
z
1
(t. T
1
. T
2
. K) =
log

P(t.T
2
)
KP(t.T
1
)

1
2

T
1
t
(v(s. T
2
) −v(s. T
1
))
2
ds

T
1
t
(v(s. T
2
) −v(s. T
1
))
2
ds
. (7.21)
Similarly, one can show that
Q
T
2
( P(T
1
. T
2
) ≥K (t)) =4(z
2
(t. T
1
. T
2
. K)) .
where
z
2
(t. T
1
. T
2
. K) =z
1
(t. T
1
. T
2
. K) +

T
1
t
(v(s. T
2
) −v(s. T
1
))
2
ds. (7.22)
Putting everything together, we obtain the following formula:
H
call
(t. T
1
. T
2
. K) =P(t. T
2
)4(z
2
(t. T
1
. T
2
. K))
−KP(t. T
1
)4(z
1
(t. T
1
. T
2
. K)) . (7.23)
for the price at time t of the call option on the zero coupon bond price. Here,
z
1
and z
2
are given by Equations (7.21) and (7.22), respectively.
7.4 Swaps and beyond in discrete time
In a discrete-time setting, we can introduce quantities similar to the ones
introduced in the continuous-time framework. However, here one has to
specify further when the actual payments take place, and there are several
ways of defining these payments. Recall the definition of the LIBOR rate
L(t. T), which is a simple rate for the interval |t. T] defined by
L(t. T) =
1−P(t. T)
(T −t)P(t. T)
. (7.24)
Similarly, the forward LIBOR rate for the interval |T

. T] is defined by
L(t. T

. T) =
P(t. T

) −P(t. T)
(T −T

)P(t. T)
. (7.25)
7.4 Swaps and beyond in discrete time 247
Swaps and swap rates
We start by describing a so-called payer interest rate swap settled in arrears
in the discrete-time setting under simple interest. Our specification involves
a set of future dates T
0
- T
1
- · · · - T
n
, where, for simplicity, we assume
that the time intervals are of equal length, i.e. that T
i
−T
i−1
= o. We refer
to T
n
as the maturity of the contract. In addition, we consider some fixed,
simple rate K and a nominal value N of the contract. The times T
1
-· · · -T
n
are the payment times and T
0
is the starting time for the measurement of the
payments. At time T
i
the holder of the contract pays the fixed amount KoN,
which is the simple interest on the nominal value N at the simple rate K
during an interval of length o. In return, the holder receives L(T
i−1
. T
i
)oN,
which is the deterministic return available at time T
i−1
for the interval from
T
i−1
to T
i
. Thus, the net payment to the holder of the contract is given by
(L(T
i−1
. T
i
) −K)oN. (7.26)
and we want to find its value at time t. To do this, we first determine the
value at time T
i−1
. This is simple, since the size of the payment is already
known at time T
i−1
. By inserting the definition of the LIBOR rate and by
discounting the payments to time T
i−1
via the zero coupon bond, we get the
value at time T
i−1
:
(1−(1+Ko)P(T
i−1
. T
i
))N. (7.27)
which at time t ≤T
i−1
has the value
(P(t. T
i−1
) −(1+Ko)P(t. T
i
))N.
This can be seen via a simple hedging argument as follows. Simply invest in
the corresponding bonds at time t and calculate the value at time T
i−1
of this
investment to see that this is identical to Equation (7.27).
By adding together the value of all these payments at times T
1
. . . . . T
n
,
we obtain the value at time t -T
0
of the payer swap as follows:
H
payer swap
(t. K) =
n
¸
i=1
(P(t. T
i−1
) −(1+Ko)P(t. T
i
))N
=N

P(t. T
0
) −P(t. T
n
) −Ko
n
¸
i=1
P(t. T
i
)

. (7.28)
If we change the signs on the cash flow, so that the holder pays at time T
i
the amount in Equation (7.26), we obtain a so-called receiver interest rate
swap settled in arrears. The value of this contract is obtained by multiplying
Equation (7.28) by −1. We can consider contracts with the payments given in
248 Interest rate derivatives in insurance
Equation (7.26) for basically any value of K; however, it is particularly
interesting to find the value of K such that the value of the contract at time t
is zero. As in the continuous-time case, this implies that two parties can enter
this agreement at time t without making any additional payments. It follows
from Equation (7.28) that this is obtained by setting K equal to
R
swap
(t. T
0
. T
n
) =
P(t. T
0
) −P(t. T
n
)
o
¸
n
i=1
P(t. T
i
)
. (7.29)
which we also refer to as the forward swap rate at time t. We note that the
forward swap rate depends on the time t as well as on the chosen set of dates
T
0
. . . . . T
n
. For t =T
0
, the formula for the forward swap rate, Equation (7.29),
reduces to
1−P(t. T
n
)
o
¸
n
i=1
P(t. T
i
)
. (7.30)
which is then simply called the the swap rate. Note the resemblance between
Equation (7.30) and the continuous-time version of the swap rate, Equa-
tion (7.6).
Typical shapes for the swap rate curves can be found in Figure 7.2.
Relations between swap rates and LIBOR forward rates
As in the continuous-time case, we can represent the swap rate as a weighted
average of forward rates. Here, in the discrete-time setting with simple interest,
0 10 20 30 40
0.00
0.02
0.04
0.06
time
e
u
r
o

s
w
a
p

r
a
t
e
Figure 7.2. Euro swap rate curves.
7.4 Swaps and beyond in discrete time 249
the relevant forward rates are the LIBOR forward rates. To obtain this repre-
sentation, we rewrite the left side of Equation (7.28) as follows:
n
¸
i=1
(P(t. T
i−1
) −(1+Ko)P(t. T
i
))N
=No
n
¸
i=1
P(t. T
i
)

P(t. T
i−1
) −P(t. T
i
)
(T
i
−T
i−1
)P(t. T
i
)
−K

=No
n
¸
i=1
P(t. T
i
) (L(t. T
i−1
. T
i
) −K) . (7.31)
where the second equality follows by using Equation (7.25), the definition of
the forward LIBOR rate with T

= T
i−1
and T = T
i
, for i = 1. . . . . n. This
shows that the value at t of the payer swap is zero provided that K is given by
¸
n
i=1
P(t. T
i
)L(t. T
i−1
. T
i
)
¸
n
i=1
P(t. T
i
)
. (7.32)
Thus, the forward swap rate can alternatively be written in the form given in
Equation (7.32), which is the discrete-time version of Equation (7.7).
Caps and floors
We have already discussed caps and floors in the continuous-time setting.
The floor is a contract which provides payments at times T
1
. . . . T
n
, where the
payment at time T
i
is of the following form:
(K−L(T
i−1
. T
i
))
+
oN. (7.33)
Thus, the holder of the floor receives the difference between the fixed level K
and the LIBOR rate, provided that this difference is positive. In this way,
the floor provides a protection against low interest rates. The contract which
only prescribes one payment of the form given in Equation (7.33) is called a
floorlet. The cap is similar to the floor, with payments (L(T
i−1
. T
i
) −K)
+
oN,
and it provides a protection against high interest rates.
By the definition of ()
+
, we have, for any x. K ∈ R, (x−K) =(x−K)
+

(K −x)
+
. Thus, the payments from the payer swap can, for any K > 0,
250 Interest rate derivatives in insurance
be expressed in terms of the payments from the cap and the floor via the
following:
(L(T
i−1
. T
i
) −K) =(L(T
i−1
. T
i
) −K)
+
−(K−L(T
i−1
. T
i
))
+
.
which shows that the payer swap is the difference between the cap and the
floor. This implies also that the value at t of the payer swap is the difference
between the values of the cap and the floor, i.e.
H
payer swap
(t. K) =H
cap
(t. K) −H
floor
(t. K).
In particular, if
K =R
swap
(0. T
0
. T
n
) =
P(0. T
0
) −P(0. T
n
)
o
¸
n
i=1
P(0. T
i
)
.
then we have, by definition of the forward swap rate, that H
payer swap
(t. K) =0
and hence H
cap
(t. K) =H
floor
(t. K). Furthermore, it follows that
H
cap
(t. K) -H
floor
(t. K)
if K >R
swap
(0. T
0
. T
n
). Similarly,
H
cap
(t. K) >H
floor
(t. K)
if K - R
swap
(0. T
0
. T
n
). Consequently, the floor at time 0 is said to be at-
the-money if K = R
swap
(0. T
0
. T
n
), in-the-money if K > R
swap
(0. T
0
. T
n
) and
out-of-the-money if K -R
swap
(0. T
0
. T
n
).
By using Equation (7.24), the definition of the LIBOR rate L(T
i−1
. T
i
), we
can rewrite the payments given in Equation (7.33) under the floor as follows:
1
P(T
i−1
. T
i
)
(1+Ko)

P(T
i−1
. T
i
) −
1
1+Ko

+
N. (7.34)
This payment takes place at time T
i
, but it is known in full already at time T
i−1
.
Thus, its value at time T
i−1
is obtained by simply multiplying Equation (7.34)
by the zero coupon bond price P(T
i−1
. T
i
), which shows that the value at time
T
i−1
of the payment at time T
i
, Equation (7.33), is given by
(1+Ko)

P(T
i−1
. T
i
) −
1
1+Ko

+
N. (7.35)
This is recognized as the payoff at time T
i−1
of (1+Ko) call options on the
zero coupon bond price P(T
i−1
. T
i
) with strike 1¡(1 +Ko). This result can
be used to obtain closed form solutions for the price of the floor, in terms of
prices on bond options.
7.4 Swaps and beyond in discrete time 251
Pricing via bond options in the Vasiˇ cek model
The alternative expression given by Equation (7.35) for the value at time T
i−1
of the payment at time T
i
allows for simple closed form solutions for the
price of the discrete-time version of caps and floors in the Vasiˇ cek model.
For the floor, the value is given by
H
V
floor
(t. K) =(1+Ko)N
n
¸
i=1
E
Q
¸
e

T
i−1
t
r(s) ds

P(T
i−1
. T
i
) −
1
1+Ko

+

(t)

=(1+Ko)N
n
¸
i=1
H
V
call

t. T
i−1
. T
i
.
1
1+Ko

.
where H
V
call
(t. T
i−1
. T
i
. 1¡(1+Ko)) is defined by Equation (7.23).
Black’s formula for caps and floors
The floor is typically priced in the market by using the so-called Black’s
formula, which can be derived by assuming that forward LIBOR rates are
lognormally distributed under a certain measure. For the payment at T
i
from
the floor (the floorlet at T
i
), Black’s formula yields the price as follows:
H
floorlet T
i
(t. K) =oNP(t. T
i
) (K4(−d
2
(T
i
. t)) −L(t. T
i−1
. T
i
)4(−J
1
(T
i
. t))) .
(7.36)
where
J
1
(T
i
. t) =
log(L(t. T
i−1
. T
i
)¡K) +
1
2
u(t)
2
(T
i−1
−t)
u(t)

T
i−1
−t
.
J
2
(T
i
. t) =J
1
(T
i
. t) −u(t)

T
i−1
−t.
For the caplet at T
i
, we obtain, similarly,
H
caplet T
i
(t. K) =oNP(t. T
i
) (L(t. T
i−1
. T
i
)4(J
1
(T
i
. t)) −K4(J
2
(T
i
. t))) .
This means that the price at time t of the floor is given by
H
floor
(t. K) =
n
¸
i=1
H
floorlet T
i
(t. K).
The crucial parameter is the volatility u(t) used at time t. The volatility
typically depends on whether we are pricing a cap or a floor and on whether
the contracts are at-the-money or not. Moreover, the volatility depends on the
maturity T
n
of the caps and floors.
252 Interest rate derivatives in insurance
Swaptions
The payer swaption gives its holder the right to enter a payer swap at a given
date, which is also called the maturity of the swaption, at a fixed rate K,
which is called the strike rate. As above, we let T
1
. . . . . T
n
be the payment
dates of the swap and denote by T
0
its reset date. The payoff of the swaption
with nominal value N and maturity T
0
thus becomes:

H
payer swap
(T
0
. K)

+
=No

n
¸
i=1
P(T
0
. T
i
) (L(T
0
. T
i−1
. T
i
) −K)

+
. (7.37)
where we have used the alternative expression, Equation (7.31), for the price at
time T
0
of the payer swap. The problem now consists in finding the value at
time t - T
0
of the contract, which pays the payoff given in Equation (7.37)
at time T
0
. This is more complicated than for the caps and floors, since the
positive part ()
+
of a sum does not split into the sum of the positive parts.
However, since by definition of the swap rate R
swap
(T
0
. T
0
. T
n
) we have that
H
payer swap
(T
0
. R
swap
(T
0
. T
0
. T
n
)) =0.
Equation (7.37) can be rewritten as follows:

H
payer swap
(T
0
. K) −H
payer swap
(T
0
. R
swap
(T
0
. T
0
. T
n
))

+
=No

n
¸
i=1
P(T
0
. T
i
)

(L(T
0
. T
0
. T
i
) −K) −(L(T
0
. T
0
. T
i
)
−R
swap
(T
0
. T
0
. T
n
))

+
=No
n
¸
i=1
P(T
0
. T
i
)

R
swap
(T
0
. T
0
. T
n
) −K

+
.
(7.38)
As with caps and floors, we say that the payer swaption is at-the-money
if K = R
swap
(t. T
0
. T
n
), in-the-money if R
swap
(t. T
0
. T
n
) > K and out-of-the-
money if we have that R
swap
(t. T
0
. T
n
) - K. This illustrates that the swap
rate R
swap
(T
0
. T
0
. T
n
) determines whether or not there is a payment from the
swaption.
Black’s formula for swaptions
Similar to the pricing of caps and floors, the market uses Black’s formula
for the pricing of swaptions. For the payer swaption, the formula reads as
follows:
No

R
swap
(t. T
0
. T
n
)4(J
1
(t)) −K4(J
2
(t))

n
¸
i=1
P(t. T
i
).
7.4 Swaps and beyond in discrete time 253
where
J
1
(t) =
log

R
swap
(t. T
0
. T
n
)¡K

+
1
2
u(t)
2
(T
0
−t)
u(t)

T
0
−t
.
J
2
(t) =J
1
(t) −u(t)

T
0
−t.
The volatility parameter u used for swaptions in the market typically differs
from the one used for the pricing of caps and floors.
Constant maturity swap (CMS) and CMS options
We end this section by describing CMS options in the current setting. We
consider times T
0
- T
1
- · · · - T
n
- T
n+1
- · · · T
n+m
, where T
1
. . . . . T
n
are
payment times and T
n+1
. . . . . T
n+m
are times used for determining the size of
the payments. We assume throughout that T
i
−T
i−1
=o. More precisely, the
payment at time T
i
, i = 1. . . . . n, is defined in terms of a certain swap rate
with a fixed maturity. As usual, we have the following swap rate:
R
swap
(T
i
. T
i
. T
i+m
) =
1−P(T
i
. T
i+m
)
o
¸
m
;=i+1
P(T
i
. T
;
)
.
The payment at time T
i
from the constant maturity swap (CMS) is now defined
as (R
swap
(T
i
. T
i
. T
i+m
) −K)o. With a CMS cap, the payments are changed to
(R
swap
(T
i
. T
i
. T
i+m
)−K)
+
o, and a CMS floor pays (K−R
swap
(T
i
. T
i
. T
i+m
))
+
o
at time T
i
, i = 1. . . . . n. Thus, the individual payments resemble the payoff
from a swaption, and hence one could suggest a Black-type formula for
valuation to obtain
H
CMS cap
(t) =o
n
¸
i=1
P(t. T
i
)

R
swap
(t. T
i
. T
i+m
)4(J
1
(t. T
i
)) −K4(J
2
(t. T
i
))

.
where
J
1
(t. T
i
) =
log

R
swap
(t. T
i
. T
i+m
)¡K

+
1
2
u(t. T
i
)
2
(T
i
−t)
u(t. T
i
)

T
i
−t
.
J
2
(t. T
i
) =J
1
(t. T
i
) −u(t. T
i
)

T
i
−t.
Similarly, the value of the CMS floor could be calculated as follows:
H
CMS floor
(t) = o
n
¸
i=1
P(t. T
i
)

K4(−J
2
(t. T
i
)) −R
swap
(t. T
i
. T
i+m
)
× 4(−J
1
(t. T
i
))) .
However, it is important to realize that the discounting for the CMS floors and
caps is completely different from the one that appears for the swaptions. In
254 Interest rate derivatives in insurance
practice, one would therefore modify these formulas by applying a so-called
convexity-correction; see, for example, Brigo and Mercurio (2001, Sect. 10.7).
7.5 A brief introduction to market models
This section briefly describes some so-called market models that can be
used to obtain Black’s formula for caps, floors and swaptions. In order to
obtain the formula for the price of the cap, one essentially needs to assume that
the forward LIBOR rates are lognormally distributed. Similarly, one assumes
that the swap rates are lognormally distributed in order to obtain Black’s
formula for the swaptions. Unfortunately, these two models are not consistent
in the sense that the LIBOR model does not lead to lognormal swap rates.
For more complete presentations of this topic, we refer again to textbooks
such as Brigo and Mercurio (2001) and Rebonato (2002).
In this section, we use again the change of numeraire techniques and
forward measures introduced in Section 7.3.
We fix times T
0
-T
1
-· · · -T
n
and assume an equal distance o between
the time points. For valuation purposes, we introduce the so-called implied
savings account, which is obtained by investing at time T
i−1
in bonds maturing
at times T
i
, etc. Using this strategy from time t to T
k
, t -T
k
, for an investment
of one unit at time t, where T
;
-t -T
;+1
, yields
B(t. T
k
) =
1
P(t. T
;+1
)
1
P(T
;+1
. T
;+2
)
· · ·
1
P(T
k−1
. T
k
)
=
1
P(t. T
;+1
)
k
¸
i=;+2
1
P(T
i−1
. T
i
)
.
(Here, the product over an empty set is defined as unity.) Thus, we can use
(B(t. T
k
))
−1
as a (stochastic) discount factor. This factor plays the same role
as the usual discount factor, exp

T
k
t
r(s) ds

, which is defined in terms of
the short rate process r. However, in the present setting, the short rate process
is not defined; we simply take as given the bonds with maturities T
1
. . . . . T
n
.
7.5.1 A market model for pricing a floorlet
In order to introduce the main ideas, we first consider the problem of pricing
a floorlet with payoff given in Equation (7.33) at time T
i
. Thus, the present
value at time t of the payment at time T
i
is given by
(B(t. T
i
))
−1
(K−L(T
i−1
. T
i
))
+
oN.
7.5 A brief introduction to market models 255
and its market price at time t is given by
E
Q
¸
(B(t. T
i
))
−1
(K−L(T
i−1
. T
i
))
+
oN (t)
¸
. (7.39)
where Q is the market measure, which has the property that all traded assets
divided by the implied savings account are Q-martingales, and where (t) is
the information available at time t. In particular, we have that
P(t. T
i
) =E
Q
¸
(B(t. T
i
))
−1
(t)
¸
.
Using the forward measure technique introduced in Section 7.3.1, we can
simplify this formula further. To see this, introduce a T
i
-forward measure
defined by
dQ
T
i
dQ
=
(B(0. T
i
))
−1
P(0. T
i
)
.
Via the usual calculations, we can now rewrite the market price of the floorlet
in terms of the T
i
-forward measure:
P(t. T
i
)E
Q
T
i
¸
(K−L(T
i−1
. T
i
))
+
oN

(t)
¸
. (7.40)
which is much more simple than the previous expression given in Equa-
tion (7.39). To calculate the conditional expected value under Q
T
i
, we only
need the conditional distribution of L(T
i−1
. T
i
) given (t) under Q
T
i
. To
obtain this, one idea is to introduce a model for the development of the
forward swap rate L(t. T
i−1
. T
i
) directly under the T
i
-forward measure Q
T
i
.
Firstly, recall that the T
i
-forward measure has the special property that the ratio
between a traded asset and the T
i
-bond prize process is a Q
T
i
-martingale. Sec-
ondly, we reexamine the definition for the forward LIBOR rate L(t. T
i−1
. T
i
),
Equation (7.25), to see that it is essentially defined as the ratio between
P(t. T
i−1
) and P(t. T
i
), which should be a Q
T
i
-martingale. Thus, one possible
model would be to assume that the dynamics for L(t. T
i−1
. T
i
) under Q
T
i
is
given by
dL(t. T
i−1
. T
i
) =\(t. T
i−1
)L(t. T
i−1
. T
i
) dW
T
i
(t). (7.41)
for t ≤T
i−1
, where W
T
i
is a standard Brownian motion under Q
T
i
and where
\(t. T
i−1
) is a deterministic function. We could assume that \ is itself a
256 Interest rate derivatives in insurance
stochastic process; however, if \ is deterministic the distribution of the
LIBOR rate,
L(T
i−1
. T
i
) =L(T
i−1
. T
i−1
. T
i
).
becomes lognormal, which allows for a simple closed form expression for
the market price of the floorlet. It follows, by using Itô’s formula, that the
solution to Equation (7.41) is given by
L(T
i−1
. T
i
) =L(t. T
i−1
. T
i
) ×
exp

T
i−1
t
\(s. T
i−1
) dW
T
i
(s) −
1
2

T
i−1
t
\(s. T
i−1
)
2
ds

.
(7.42)
Since we have assumed that \ is a deterministic function, this shows that the
conditional distribution under Q
T
i
of log(L(T
i−1
. T
i
)), given (t), is normal,
with variance given by
l(t. T
i−1
) :=

T
i−1
t
\(s. T
i−1
)
2
ds.
and mean given by
log(L(t. T
i−1
. T
i
)) −
1
2
l(t. T
i−1
).
Thus, calculations similar to the ones used for the derivation of the Black–
Scholes formula show that the price of the floorlet, Equation (7.40), can be
written as follows:
¨
H
floorlet T
i
(t. K) =P(t. T
i
)

K4(−
¨
J
2
(t. T
i−1
))
−L(t. T
i−1
. T
i
)4(−
¨
J
1
(t. T
i−1
))

. (7.43)
where
¨
J
1
(t. T
i−1
) =
log

L(t.T
i−1
.T
i
)
K

+
1
2
l(t. T
i−1
)

l(t. T
i−1
)
and
¨
J
2
(t. T
i−1
) =
¨
J
1
(t. T
i−1
) −
1
2

l(t. T
i−1
).
It is worth mentioning that the valuation formula, Equation (7.43), differs from
the market formula, Equation (7.36), only via the variance term l(t. T
i−1
). In
fact, if we choose \(t. T
i−1
) independent of T
i−1
, then the two formulas are
identical.
7.7 A portfolio of contracts 257
The floor consisting of floorlets at times T
1
. . . . . T
n
can now be valued by
modelling forward LIBOR rates L(t. T
i−1
. T
i
) for any i =1. . . . . n under the
corresponding forward measures Q
T
1
. . . . . Q
T
n
.
As mentioned above, Black’s formula for swaptions cannot be obtained
within this model, since swap rates are not log-normally distributed.
7.6 Interest rate derivatives in insurance
We recall the basic example introduced in Chapter 2 of an endowment with a
continuously paid premium. It was demonstrated in Chapter 3 how the guar-
anteed payments from this contract could (in principle) be hedged perfectly by
investing in zero coupon bonds. Here we address the more realistic situation,
where the insurance company does not have access to the zero coupon bonds
that are necessary for this perfect hedge. In this situation, the market for long
term interest rate derivatives provides additional possibilities for reducing the
risk inherent in the insurance contract.
Here, one has to be a little more precise and specify in more detail the
optimization criteria used by the company. There are several different possi-
bilities, which we describe vaguely here.
One idea is to require that the total reserves at any time exceed the
market value of the guaranteed payments. This can be viewed as the minimal
requirement, i.e. this condition cannot be relaxed. One of the main current
problems to life insurers seems to be handling of policies with a high first
order rate, which have been issued in periods with very high (nominal) rates
and which have not been fully hedged. Firstly, the guarantees were originally
not taken seriously, since the guarantees (of say 4.5%) were substantially
lower than the nominal rates (say 15%). Secondly, a perfect hedge was not
really possible in the bond market, due to the extremely long time horizons
associated with life and pension insurance contracts. Today, these contracts
represent a major risk to life insurance companies and pension funds, since
they essentially guarantee policy holders yearly returns which exceed current
short rates.
7.7 A portfolio of contracts
We analyze the impact of interest rate derivatives by considering some simple
examples. We restrict the analysis to the continuous-time framework presented
in Section 7.2.
258 Interest rate derivatives in insurance
Pure endowment repeated
Consider as a simple example the pure endowment paid by single pre-
mium, which was also analyzed as an introductory example in Chapters 3 and
5. We assume that premiums are fixed using some deterministic interest rate
r

, which is chosen on the safe side, such that the premium payable at time
0 for a T-year pure endowment contracts with sum insured 1 is given by
r(0) =
T
¡
x
e

T
0
r

(s) ds
.
which can also be written in terms of a decrement series by using the formula
T
¡
x
=/
x+T
¡/
x
; see Section 3.2.4 for more details. As in the previous chapters,
we consider a portfolio of /
x
identical contracts, such that the present value
at time 0 of benefits less premiums is given by
/
x+T
e

T
0
r(s) ds
−/
x
r(0).
Note that we are here discounting the payments by using the market interest
rate r, which is a stochastic process; one possibility is to use the Vasiˇ cek
model for r, which was discussed in Chapter 3. As mentioned in Chapter 3,
the insurer’s liability can be hedged perfectly by investing in a certain amount
of T-bonds (zero coupon bonds). Since this bond might not be available in the
market, we investigate here the effect of using a swap instead of the T-bond.
More precisely, we assume that the company invests premiums in the savings
account (which we can think of as the result of investing in bonds with very
short time to maturity). If we put the premiums into an account which bears
interest r, the value at time t becomes
H
premium
(t) =/
x
r(0) e

t
0
r(s) ds
.
Finally, the market value of the guaranteed future payments at time t is
given by
V
g
(t) =/
x+t T−t
¡
x+t
P(t. T). (7.44)
where we have assumed that the market mortality intensity is equal to the
mortality intensity used for calculation of the premium.
7.7.1 Hedging via a static swap position
The swap account
We assume now that the company enters N =/
x
r(0)(1+a) receiver swaps
with strike «, which in the continuous-time formulation means that the com-
pany receives fixed interest « and pays floating interest r. (We refer to N as
7.7 A portfolio of contracts 259
the face value.) The present value at time 0 of the payments from one unit of
the swap is given by

T
0
e

t
0
r(s)
(«−r(t)) dt. (7.45)
and we assume that the strike « is the T-year swap rate at time 0, «(0. T),
which is defined by Equation (7.6). This means that the company can enter the
contract at time 0 without paying anything to the other party. The disadvantage
of this contract is, of course, that the company has to pay coupons (r(t) −«)
to the other party if the interest exceeds the strike «. Let us imagine that we
create an account in the company which consists of the market value of the
swap to which are added past incomes and from which are subtracted past
outgoes, and that everything is accumulated with the market interest r. We
denote by H
swap
(t) the value of this account at time t. Since « =«(0. T), we
have H
swap
(0) =0. At time t, the value is given by
H
swap
(t) =N

t
0
e

t
t
r(s) ds
(«−r(t)) dt
+NE
Q
¸

T
t
e

t
t
r(s)
(«−r(t)) dt (t)
¸
.
which is past incomes less outgoes added to the market value of future
payments. By using calculations similar to the ones following Equation (7.3),
we can rewrite the first term on the right as follows:

t
0
e

t
t
r(s) ds
dt −N(e

t
0
r(s) ds
−1).
For the second part, we obtain

T
t
P(t. t) dt −N(1−P(t. T)).
Putting everything together, we obtain
H
swap
(t) =N«

t
0
e

t
t
r(s) ds
dt +

T
t
P(t. t) dt

−N(e

t
0
r(s) ds
−P(t. T)).
The total account
We can now compare the value of the investments (the accounts for the paid
premiums and the swap account) with the market value of the liabilities. Thus,
260 Interest rate derivatives in insurance
we study the process defined as the difference between these quantities in
more detail, and let
U(t) =H
premium
(t) +H
swap
(t) −V
g
(t)
=/
x
r(0) e

t
0
r(s) ds
+N«

t
0
e

t
t
r(s) ds
dt +

T
t
P(t. t) dt

−N

e

t
0
r(s) ds
−P(t. T)

−/
x+t T−t
¡
x+t
P(t. T).
The investments should ensure that assets exceed liabilities, i.e. it is desirable
that this account is non-negative. Note that if we take a = 0, such that
N =/
x
r(0), we obtain the following more simple equation:
U(t) =/
x
r(0)P(t. T) −/
x+t T−t
¡
x+t
P(t. T)
+/
x
r(0) «

t
0
e

t
t
r(s) ds
dt +

T
t
P(t. t) dt

=«/
x+T
e

T
0
r

(s) ds

t
0
e

t
t
r(s) ds
dt +

T
t
P(t. t) dt

−/
x+T
P(t. T)

1−e

T
0
r

(s) ds

.
where, in particular,
U(0) =/
x
r(0) −/
xT
¡
x
P(0. T) =/
x+T

e

T
0
r

(s) ds
−P(0. T)

.
which is recognized as the individual bonus potential at time t; see Sec-
tion 3.3.5. At time T, the value of the total account is given by
U(T) =/
x+T

«e

T
0
r

(s) ds

T
0
e

T
t
r(s) ds
dt −

1−e

T
0
r

(s) ds

.
It is interesting to compare the distribution of U(t) with that of H
premium
(t) −
V
g
(t), which corresponds to the situation where the company has not entered
any swaps. In addition, one can study the sensitivity of U(t) to changes in
the current short rate r(t). A more refined version of the present study is
to modify the company’s investment strategy such that the company invests
premiums in bonds with a longer maturity.
We present here a simple simulation study and compare the distribution
of U(T) with that of
H
premium
(T) −V
g
(T−) =/
x+T

e

T
0
(r(s)−r

(s)) ds
−1

.
7.7 A portfolio of contracts 261
which is the terminal value of the account if we are not investing in any
swaps, but only using the savings account (a roll-over strategy in bonds with
very short maturity).
Numerical example
We present some simulation results for the Vasiˇ cek model with the parameters
r(0) = 0.03, u = 0.03, o = 0.36 and l¡o = 0.06, which were also used in
Chapter 3 and in Section 7.3.1. We take /
x+T
=1.
Figures 7.3 and 7.4 show histograms for the terminal values of the total
accounts without and with swaps, respectively. The figures indicate how the
swap can be used to center the distribution of the terminal value of the
account, so that the impact of very low interest rates is reduced. The cost is
that in situations with high interest, the account is reduced with the swap,
account without swap
0.2 0.4 0.6 0.8
7
6
5
4
3
2
1
0
Figure 7.3. Histogram for the total account without swaps.
account with swap
0.2 0.4 0.6 0.8
7
6
5
4
3
2
1
0
Figure 7.4. Histogram for the total account with swaps.
262 Interest rate derivatives in insurance
since the insurance company has to make a net payment when the short rate
exceeds the swap rate «.
We end the analysis here. Obviously, these calculations could be performed
for the other types of interest rate derivatives described in this chapter. Simi-
larly, one could include the mortality aspect and introduce optimality criteria
for determining the optimal number of interest rate derivatives that should be
included in order to control the combined mortality and financial risk inherent
in these contracts.
Appendix
A.1 Some results from probability theory
We recall some well known concepts from probability theory that might be
helpful when reading the introduction to arbitrage theory.
Probability spaces and random variables
When we talk about random variables, X and Y, say, they are typically defined
on some probability space (D. . P); D is a set which can be interpreted as
all states of the world; is a u-algebra of subsets of D; P is a probability
measure, which, in particular, describes the joint distribution of the pair (X. Y).
The measure P will be called the true measure or the physical measure, since
it describes the true nature of the random variables. We use the phrase almost
surely (abbreviated a.s. or P-a.s.) when describing properties for random
variables which hold for (almost) all w∈ D, except for w’s in some negligible
subset N ∈ with P(N) = 0. For example, we say that “X is constant a.s.”
if there exists some real number K and some set N ∈ with P(N) =0 such
that X =K for all w ∈ D\N.
Random phenomena over time
We are interested in modeling the evolution of bond prices and other prices
over a certain fixed time interval |0. T]. For example, it is natural to require
that the price P(t. t) at time t of a zero coupon bond with maturity t,
see Chapter 3, is a random variable which is not revealed before time t.
The family of prices up to time T, (P(t. t))
t∈|0.T]
, is an example of a stochastic
process. More generally, a stochastic process is a family of random variables
X =(X(t))
t∈|0.T]
. In order to model more precisely what is known and what is
263
264 Appendix
not known at time t, one introduces a so-called filtration F, which describes
the amount of information available at any time during the time interval |0. T].
Mathematically, a filtration is an increasing sequence of sub-u-algebras of
, also written F =((t))
t∈|0.T]
, indexed by time t. This means that (s) ⊆
(t) ⊆ for any s -t ≤T. The u-algebra (t) includes all the information
that is available at time t. The stochastic process X is now said to be adapted
(to the filtration F) if X(t) can be determined from the u-algebra (t) for
any t ∈ |0. T], i.e. if X(t) is observed at time t; mathematically this means
that X(t) is (t)-measurable. An adapted process is called predictable if the
process X satisfies some additional measurability conditions. In particular, an
adapted process X is predictable if it is deterministic (completely known at
time 0) or if it is left-continuous. In discrete time, the concept of predictability
is more simple; see Section 3.7.
Martingales
A stochastic process X is said to be a martingale if for any t -u we have that
E|X(u)(t)] =X(t). In particular, if X is a Markov process with respect to
the filtration F, then the condition simplifies to E|X(u)X(t)] = X(t). This
has the following interpretation. If X(t) and X(u) represent the price of a
bond at time t (today) and u (tomorrow), respectively, then we see that, if X
is a martingale, then the average price for tomorrow, given today’s price, is
exactly equal to the price today.
Change of measure
It will be necessary to introduce probability measures that are different from
the true measure, or, in other words, to change the probability measure. For
any non-negative random variable Z ≥ 0 with E
P
|Z] =

D
Z(w) dP(w) = 1,
we can define a new probability measure Q as follows:
Q(A) =E
P
|1
A
Z] =

A
Z(w) dP(w). (A.1)
for A∈ . In particular, if Z >0P-a.s., then ∀A∈ : P(A) =0 ⇔Q(A) =0.
In this case, P and Q are said to be equivalent, and Z is called the density for
Q with respect to P.
Change of measure for remaining lifetimes
Actuaries in life insurance are in fact used to work with several different
equivalent probability measures at the same time! Consider, for example,
a group of insured individuals whose true mortality intensity is given by
A.1 Some results from probability theory 265
µ(x+t), i.e. this is the mortality intensity under the true probability measure
P. The true survival probability for an individual with remaining lifetime T
x
is then given by
P(T
x
>t) =exp

t
0
µ(x+u) du

=
t
¡
x
.
For the computation of premiums, actuaries would then introduce a first order
mortality intensity µ

(t +x), which means that, under some other probability
measure P

, the survival probability is P

(T
x
>t) =exp(−

t
0
µ

(x+u) du) =
t
¡

x
. Provided that µ and µ

are sufficiently regular and provided that µ

(x+t)
is only equal to zero if and only if µ(x+t) is, then one can actually define
P

via Equation (A.1) by using the following:
Z =exp

T
0


(x+s) −µ(x+s))1
]T
x
>s]
ds

µ

(x+T
x
)
µ(x+T
x
)

1
]T
x
-T]
.
(A.2)
Since this type of argument is not given in typical textbooks on life insurance
mathematics, we give the proof in some detail here. It should be proved
that E
P
|Z] = 1 and that the measure P

defined by dP

¡dP = Z satisfies
P

(T
x
> t) = exp(−

t
0
µ

(x+u) du). It follows immediately by integrating
with respect to the distribution of T
x
, whose density is given by
t
¡
x
µ(x+t),
that
E
P
|Z] =

T
0
t
¡
x
µ(x+t)

e

t
0


(x+s)−µ(x+s)) ds
µ

(x+t)
µ(x+t)

dt
+

T
t
¡
x
µ(x+t)

e

T
0


(x+s)−µ(x+s)) ds

dt
=

T
0
e

t
0
µ

(x+s) ds
µ

(x+t) dt
+e

T
0


(x+s)−µ(x+s)) ds

T
t
¡
x
µ(x+t) dt
=(1−e

T
0
µ

(x+s) ds
) +e

T
0


(x+s)−µ(x+s)) ds
e

T
0
µ(x+s) ds
=1.
266 Appendix
For t -T, we find from similar calculations the following:
P

(T
x
>t) =E
P
¸
1
]T
x
>t]
Z
¸
=

T
t
u
¡
x
µ(x+u)

e

u
0


(x+s)−µ(x+s)) ds
µ

(x+u)
µ(x+u)

du
+

T
u
¡
x
µ(x+u)

e

T
0


(x+s)−µ(x+s)) ds

du
=e

t
0
µ

(x+s) ds
=
t
¡

x
.
so that µ

is indeed the mortality intensity under the measure P

with density
given in Equation (A.2). More on this subject can be found in Møller (1998)
and Steffensen (2000).
A.2 The risk-minimizing strategy
In this appendix we give the direct construction of the risk-minimizing strategy
outlined in Equations (5.56) and (5.57) proposed by Föllmer and Schweizer
(1988). The cost process is defined by
C(t. h) =V(t. h) −
t
¸
;=1
h
1
(;)AX(;). (A.3)
and we consider the problem of minimizing:
r(t. h) =E
Q
|(C(t +1. h) −C(t. h))
2
(t)]. (A.4)
considered as a function of (h
1
(t +1). h
0
(t)) under the side condition V(T. h) =
H. Since H is assumed to be (T)-measurable, we see from the decompo-
sition in Equation (5.55) that
H =E
Q
|H (T)] =V

(0) +
T
¸
;=1
h
1
(;. H)AX(;) +L(T. H). (A.5)
It now follows via standard arguments, similar to those given in Section 5.5,
that the minimum for Equation (A.4) is obtained for a strategy
¯
h, where
¯
h
0
is chosen such that
C(t.
¯
h) =E
Q
|C(t +1.
¯
h)(t)].
i.e. such that C(
¯
h) is a Q-martingale. By using Equation (A.3), we see that the
value process V(
¯
h) is also a Q-martingale. Since, furthermore, V(T.
¯
h) =H,
A.3 Risk minimization for unit-linked contracts 267
Equation (A.5) shows that
V(t.
¯
h) =V

(0) +
t
¸
;=1
h
1
(;. H)AX(;) +L(t. H). (A.6)
and hence
¯
h
0
(t) = V

(t) −
¯
h
1
(t)X(t). This fixes
¯
h
0
in terms of
¯
h
1
. Our next
step is to insert Equations (A.3) and (A.6) into Equation (A.4), which leads
to the following:
r(t.
¯
h) =E
Q
¸

(h
1
(t +1. H) −
¯
h
1
(t +1))AX(t +1) +AL(t +1. H)

2

(t)
¸
.
(A.7)
Since
¯
h
1
(t +1) and h
1
(t +1. H) are (t)-measurable, and since
E
Q
|AX(t +1)AL(t +1. H) (t)] =0.
by the orthogonality between X and L
H
, we can rewrite Equation (A.7) as
follows:
(h
1
(t +1. H) −
¯
h
1
(t +1))
2
E
Q
¸
(AX(t +1))
2

(t)
¸
+E
Q
¸
(AL(t +1. H))
2

(t)
¸
.
which is minimized for
¯
h
1
(t +1) = h
1
(t +1. H). This shows that the risk-
minimizing strategy is given by
(
¨
h
1
(t).
¨
h
0
(t)) =(h
1
(t. H). V

(t) −h
1
(t. H)X(t)).
A.3 Risk minimization for unit-linked contracts
Since Y and S are assumed to be independent, we obtain the following:
V

(t) : =E
Q
¸
Y(T)
1(S
1
(T))
S
0
(T)

(t)
¸
=E
Q
¸
1(S
1
(T))
S
0
(T)

(t)
¸
E
Q
| Y(T) (t)]
=E
Q
¸
1(S
1
(T))
S
0
(T)

(t)
¸
E
Q
| Y(T) (t)]
=r(t. 1)M(t).
In the second equality, we have used the independence between Y(T) and
S
1
(T), and in the third equality we have used the definition of the u-algebra
268 Appendix
(t) as well as the independence between the two sources of risk. The last
equality follows from the definitions of r(1) and M. Now, note that
AV

(t) =V

(t) −V

(t −1) =r(t. 1)M(t) −r(t −1. 1)M(t −1)
=(r(t. 1) −r(t −1. 1))M(t −1) +r(t. 1)(M(t) −M(t −1))
=M(t −1)o(t. 1)AX(t) +r(t. 1)AM(t).
where we have used the representation given in Equation (5.47) for r(1) in
the last equality. The term M(t −1)o(t. 1) is (t −1)-measurable, such that
the process h
1
(H) defined by
h
1
(t. H) =M(t −1)o(t. 1)
is predictable. If we can show that the process L(H) defined by
L(t. H) =
t
¸
;=1
r(;. 1)AM(;)
is a Q-martingale, and that the product L(H)X is also a Q-martingale, then
we have actually verified that the decomposition in Equation (5.55) is indeed
given by
V

(t) =V

(0) +
t
¸
;=1
M(; −1)o
1
(;)AX(;) +
t
¸
;=1
r(;. 1)AM(;). (A.8)
To see that L(H) is a martingale, we apply the rule of iterated expectations
and the independence between the two sources of risk, or, more precisely,
between the two filtrations G and H. Using these ingredients, we obtain
E
Q
| AL(t. H) (t −1)] =E
Q
| r(t. 1)AM(t) (t −1)]
=E
Q
¸
r(t. 1)E
Q
| AM(t)(t) ∨(t −1)]

(t −1)
¸
=0.
since M is a martingale and since M is independent of the filtration G. Similar
calculations show that L(H)X is also a martingale. Since
A(L(H)X)(t) =L(t. H)X(t) −L(t −1. H)X(t −1)
=AL(t. H)AX(t) +L(t −1. H)AX(t) +X(t −1)AL(t. H).
it is sufficient to show that
E
Q
| AL(t. H)AX(t) (t −1)] =0.
A.4 Mean-variance hedging for unit-linked contracts 269
This now follows by using calculations similar to the ones used to show that
L(H) is a martingale, so that Equation (A.8) is indeed the desired decomposi-
tion. From the results in Section A.2 it now follows that the risk-minimizing
strategy is as follows:
h
1
(t) =Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1).
h
0
(t) =Y(t)
T−t
¡
x+t
r(t. 1) −Y(t −1)
T−(t−1)
¡
x+(t−1)
o(t. 1)X(t).
A.4 Mean-variance hedging for unit-linked contracts
In this section, we derive the mean-variance hedging strategy of Proposi-
tion 5.8 for the unit-linked contract given in Equation (5.46). Firstly, we
examine the structure of the probability measures
¯
P. In the binomial model,
the increments of the discounted stock price process are given by
AX(t) =X(t −1)
Z(t) −r
1+r
.
and hence the (conditional) expected return is as follows:
AA(t) =E|AX(t)(t −1)] =X(t −1)
E|Z(t)] −r
1+r
.
Similarly, the (conditional) variance on the return in year t is given by
Var|AX(t)(t −1)] =
X(t −1)
2
(1+r)
2
Var|Z(t) −r].
In addition, we have that
¯
\(t) =
AA(t)
E|(AX(t))
2
(t −1)]
=
X(t −1)
E|Z(t)]−r
1+r
X(t −1)
2
E
¸

Z(t)−r
1+r

2
¸
=
1+r
X(t −1)
E|Z(t)] −r
E|(Z(t) −r)
2
]
.
so that
¯
\(t)AA(t) is deterministic. We can now describe more precisely the
properties of
¯
P. In particular, we see that
¯
P is indeed a martingale measure,
i.e. X is a
¯
P-martingale. To see this, note that
1−
¯
\(t)AX(t)
1−
¯
\(t)AA(t)
=
E|(Z(t) −r)
2
] −(Z(t) −r)(E|Z(t)] −r)
E|(Z(t) −r)
2
] −(E|Z(t)] −r)
2
.
270 Appendix
such that the factors in the definition of
¯
P are independent. One can use this
independence to show that Z(1) . . . . . Z(t) are i.i.d. under
¯
P, and that
¯
E|AX(t)(t −1)] =E
¸
d
¯
P
dP
AX(t)

(t −1)

=0. (A.9)
Thus,
¯
P is actually identical to the probability measure Q that was already
introduced in Section 5.7.1.
It is now relatively easy to show that the decomposition given in Equa-
tion (5.71) is identical to the representation given in Equation (5.63) which
was derived for the risk-minimizing strategy. The proposition now follows
directly by using Equation (5.70) and by choosing the deposit
¯
h
0
in the sav-
ings account such that the strategy is self-financing. For completeness, these
calculations are included here.
In Section 5.7.5, we worked with the martingale measure Q which was
obtained as the product measure of the unique martingale measure for the
binomial market and the physical measure for the insured lifetimes. Under
this measure, we derived the Kunita–Watanabe decomposition for the Q-
martingale V

(t) =E
Q
|H(t)] and showed that it was given by
V

(t) =V

(0) +
t
¸
;=1
h
1
(;. H)AX(;) +L(t. H).
where
h
1
(;. H) =M(; −1)o(;. 1).
L(t. H) =
t
¸
;=1
r(;. 1)AM(;).
Here, L(H) is a Q-martingale which is orthogonal to the Q-martingale, i.e.
E
Q
| L(u. H)X(u) (t)] =L(t. H)X(t)
for all u ≥ t. In order to solve the mean-variance hedging problem, we need
the decomposition from Equation (5.71) given by
H =
¯
H(0) +
T
¸
;=1
¯
h
1
(;. H)AX(;) +
¯
L(T. H). (A.10)
where
¯
L(H) is a P-martingale which is orthogonal to the P-martingale X−A.
This implies that one cannot conclude directly that the decomposition in
Equation (A.10) can be obtained from the Kunita–Watanabe decomposition
A.4 Mean-variance hedging for unit-linked contracts 271
for the Q-martingale V

. However, we claim that the decomposition in Equa-
tion (A.10) is given by
¯
H(0) =V

(0).
¯
h
1
(t. H) =h
1
(t. H).
¯
L(t. H) =L(t. H).
To see this, one has to check that
¯
L(H) is a P-martingale (we only know that
it is a Q-martingale) and that
¯
L(H) is orthogonal to X−A. We note that
¯
L(t. H) =
t
¸
;=1
r(;. 1)AM(;)
is indeed a P-martingale, since M is a P-martingale and since r(;. 1) is
stochastically independent of AM(;). To see that
¯
L(H) is orthogonal to X−A,
it is sufficient to verify that
E| (AX(t) −AA(t))AL(t. H) (t −1)] =0.
and this follows via calculations similar to the ones used in Section A.3.
By conditioning on (t) ∨(t −1) and by using the independence between
(T) and (T) and the fact that M is a P-martingale, we see that
E
¸
(AX(t) −AA(t))A
¯
L(H. t)

(t −1)
¸
=E| (AX(t) −AA(t))E| r(t. 1)AM(t) (t) ∨(t −1)] (t −1)]
=E| (AX(t) −AA(t))r(t. 1)E| AM(t) (t) ∨(t −1)] (t −1)]
=0.
Here, the last equality follows by exploiting the fact that M is a P-martingale.
References
Aase, K. K. and Persson, S. A. (1994). Pricing of unit-linked life insurance policies,
Scandinavian Actuarial Journal, pp. 26–52.
Bacinello, A. R. (2001). Fair pricing of life insurance participating policies with a
minimum interest rate guaranteed, ASTIN Bulletin 31, 275–297.
Bacinello, A. R. and Ortu, F. (1993a). Pricing equity-linked life insurance with endoge-
nous minimum guarantees, Insurance: Mathematics and Economics 12, 245–257.
Bacinello, A. R. and Ortu, F. (1993b). Pricing Guaranteed Securities-Linked Life
Insurance under Interest-Rate Risk. Proceedings of the 3rd AFIR Colloquium,
Rome, Vol. I, pp. 35–55.
Bacinello, A. R. and Persson, S. A. (1998). Design and Pricing of Equity-Linked
Life Insurance under Stochastic Interest Rates. Preprint, Institute of Finance
and Management Science, The Norwegian School of Economics and Business
Administration.
Baxter, M. and Rennie, A. (1996). Financial Calculus. An Introduction to Derivative
Pricing (Cambridge: Cambridge University Press).
Becherer, D. (2003). Rational hedging and valuation of integrated risks under constant
absolute risk aversion, Insurance: Mathematics and Economics 33, 1–28.
Bielecki, T. R. and Rutkowski, M. (2001). Credit Risk: Modelling, Valuation and
Hedging (Berlin: Springer).
Biffis, E. and Millossovich, P. (2006). The fair value of guaranteed annuity options,
Scandinavian Actuarial Journal, pp. 23–41.
Björk, T. (1994). Stokastisk Kalkyl och Kapitalmarknadsteori. Lecture notes, Kungl.
Tekniska Högskolan, Stockholm.
Björk, T. (1997). Interest rate theory, in Runggaldier, W. J. (ed.), Financial Math-
ematics, Springer Lecture Notes in Mathematics, Vol. 1656 (Berlin: Springer
Verlag).
Björk, T. (2004). Arbitrage Theory in Continuous Time, 2nd edn (Oxford: Oxford
University Press).
Black, F. and Scholes, M. (1973). The pricing of options and corporate liabilities,
Journal of Political Economy 81, 637–654.
Blake, D. (1998). Pension schemes as options on pension fund assets: implications for
pension fund management, Insurance: Mathematics and Economics 23, 263–286.
272
References 273
Blake, D., Cairns, A. J. G. and Dowd, K. (2006). Living with mortality: longevity bonds
and other mortality-linked securities. Preprint, Actuarial Mathematics and Statis-
tics, School of Mathematical and Computer Sciences, Heriot-Watt University,
Edinburgh.
Brennan, M. J. and Schwartz, E. S. (1976). The pricing of equity-linked life insur-
ance policies with an asset value guarantee, Journal of Financial Economics 3,
195–213.
Brennan, M. J. and Schwartz, E. S. (1979a). Alternative investment strategies for the
issuers of equity-linked life insurance with an asset value guarantee, Journal of
Business 52, 63–93.
Brennan, M. J. and Schwartz, E. S. (1979b). Pricing and Investment Strategies
for Guaranteed Equity-Linked Life Insurance. Monograph no. 7. The S. S.
Huebner Foundation for Insurance Education, Wharton School, University of
Pennsylvania, Philadelphia.
Brigo, D. and Mercurio, F. (2001). Interest Rate Models: Theory and Practice, (Berlin:
Springer).
Briys, E. and de Varenne, F. (1994). Life insurance in a contingent claim framework:
pricing and regulatory implications, The Geneva Papers on Risk and Insurance
Theory 19, 53–72.
Briys, E. and de Varenne, F. (1997). On the risk of life insurance liabili-
ties: debunking some common pitfalls, Journal of Risk and Insurance 64,
673–694.
Cairns, A. J. G. (2000). Some notes on the dynamics and optimal control
of stochastic pension fund models in continuous time, ASTIN Bulletin 30,
19–55.
Cairns, A. J. G. (2004). Interest Rate Models: An Introduction (Princeton: Princeton
University Press).
Cairns, A. J. G., Blake, D. and Dowd, K. (2006). Pricing death: frameworks for the
valuation and securitization of mortality risk. ASTIN Bulletin 36, 79–120.
Cont, R. and Tankov, P. (2003). Financial Modelling with Jump Processes (Boca
Raton: Chapman & Hall/CRC Press).
Cowley, A. and Cummins, J. D. (2005). Securitization of life insurance assets and
liabilities, The Journal of Risk and Insurance 72, 193–226.
Cox, J., Ingersoll, J. and Ross, S. (1985). A theory of the term-structure of interest
rates, Econometrica 53, 385–408.
Cox, J., Ross, S. and Rubinstein, M. (1979). Option pricing: a simplified approach,
Journal of Financial Economics 7, 229–263.
Dahl, M. (2004). Stochastic mortality in life insurance: market reserves and mortality-
linked insurance contracts, Insurance: Mathematics and Economics 35, 113–136.
Dahl, M. and Møller, T. (2006). Valuation and Hedging of Life Insurance Liabilities
with Systematic Mortality Risk, Insurance: Mathematics and Economics, 39,
193–217.
Dana, R. A. (1999). Existence, uniqueness and determinacy of equilibrium in C.A.P.M.
with a riskless asset, Journal of Mathematical Economics 32, 167–175.
Delbaen, F. (1986). Equity linked policies, Bulletin de l’Association Royale des Actu-
aires Belges, pp. 33–52.
274 References
Delbaen, F. and Haezendonck, J. (1989). A martingale approach to premium calculation
principles in an arbitrage-free market, Insurance: Mathematics and Economics
8, 269–277.
El Karoui, N. and Quenez, M. C. (1995). Dynamic programming and pricing of contin-
gent claims in an incomplete market, SIAM Journal on Control and Optimization
33, 29–66.
Embrechts, P., Klüppelberg, C. and Mikosch, T. (1997). Modelling Extremal Events in
Insurance and Finance (Berlin: Springer-Verlag).
Föllmer, H. and Leukert, P. (1999). Quantile hedging, Finance and Stochastics
3, 251–273.
Föllmer, H. and Schied, A. (2002). Stochastic Finance: An Introduction in Discrete
Time, de Gruyter Series in Mathematics 27 (Berlin: Walter de Gruyter).
Föllmer, H. and Schweizer, M. (1988). Hedging by sequential regression: an introduc-
tion to the mathematics of option trading, ASTIN Bulletin 18, 147–160.
Föllmer, H. and Sondermann, D. (1986). Hedging of non-redundant contingent claims,
in W. Hildenbrand and A. Mas-Colell (eds), Contributions to Mathematical
Economics (Amsterdam: North-Holland), pp. 205–223.
Gerber, H. U. (1997). Life Insurance Mathematics (Berlin: Springer-Verlag).
Grosen, A. and Jørgensen, P. L. (1997). Valuation of early exercisable interest rate
guarantees, Journal of Risk and Insurance 64, 481–503.
Grosen, A. and Jørgensen, P. L. (2000). Fair valuation of life insurance liabilities:
the impact of interest rate guarantees, surrender options, and bonus policies,
Insurance: Mathematics and Economics 26, 37–57.
Grosen, A. and Jørgensen, P. L. (2002). Life insurance liabilities at market values: an
analysis of insolvency risk, bonus policy, and regulatory intervention rules in a
barrier option framework, Journal of Risk and Insurance 69, 63–91.
Hansen, M. and Miltersen, K. (2002). Minimum rate of return guarantees: the Danish
case, Scandinavian Actuarial Journal, pp. 280–318.
Hardy, M. (2003). Investment Guarantees: Modeling and Risk Management for Equity-
Linked Life Insurance (Hoboken, NJ: Wiley).
Harrison, J. M. and Kreps, D. M. (1979). Martingales and arbitrage in multiperiod
securities markets, Journal of Economic Theory 20, 381–408.
Harrison, J. M. and Pliska, S. R. (1981). Martingales and stochastic integrals in the
theory of continuous trading, Stochastic Processes and their Applications 11,
215–260.
Heston, S. L. (1993). A closed-form solution for options with stochastic volatility
with applications to bond and currency options, The Review of Financial Studies
6, 327–343.
Hoem, J. (1969). Markov chain models in life insurance, Blätter der Deutschen
Gesellschaft für Versicherungsmathematik 9, 91–107.
Hull, J. C. (2005). Options, Futures, and other Derivatives, 6th edn. (Upper Saddle
River, NJ: Prentice-Hall).
Jarrow, R. (1996). Modelling Fixed Income Securities and Interest Rate Options
(New York: McGraw-Hill).
Jørgensen, P. L. (2004). On accounting standards and fair valuation of life insurance
and pension liabilities, Scandinavian Actuarial Journal, pp. 372–394.
References 275
Kahn, P. M. (1971). Projections of variable life insurance operations, with discussion,
Transactions of the Society of Actuaries, XXIII, 335–366.
Kalashnikov, V. and Norberg, R. (2003). On the sensitivity of premiums and reserves
to changes in valuation elements, Scandinavian Actuarial Journal, pp. 238–256.
Lamberton, D. and Lapeyre, B. (1996). Introduction to Stochastic Calculus Applied to
Finance (Boca Raton: Chapman & Hall).
Lando, D. (2004). Credit Risk Modeling – Theory and Applications (Princeton: Princeton
University Press).
Lee, R. D. (2000). The Lee–Carter method for forecasting mortality, with various
extensions and application, North American Actuarial Journal 4, 80–93.
Lee, R. D. and Carter, L. (1992). Modeling and forecasting the time series of US
mortality, Journal of the American Statistical Association 87, 659–671.
Marocco, P. and Pitacco, E. (1998). Longevity Risk and Life Annuity Reinsurance.
Transactions of the 26th International Congress of Actuaries, Birmingham,
England, Vol. 6, 453–479.
Merton, R. C. (1973). Theory of rational option pricing, Bell Journal of Economics
and Management Science 4, 141–183.
Milevsky, M. A. and Promislow, S. D. (2001). Mortality derivatives and the option to
annuitise, Insurance: Mathematics and Economics 29, 299–318.
Miltersen, K. R. and Persson, S. A. (2000). A Note on Interest Rate Guarantees and
Bonus: The Norwegian Case. Proceedings of the 10th AFIR Colloquium, Tromsø,
pp. 507–516.
Miltersen, K. R. and Persson, S. A. (2003). Guaranteed investment contracts: distributed
and undistributed excess return, Scandinavian Actuarial Journal, pp. 257–279.
Møller, T. (1998). Risk-minimizing hedging strategies for unit-linked life insurance
contracts, ASTIN Bulletin 28, 17–47.
Møller, T. (2000). Quadratic hedging approaches and indifference pricing in insurance,
Ph.D. thesis, University of Copenhagen.
Møller, T. (2001a). Hedging equity-linked life insurance contracts, North American
Actuarial Journal 5, 79–95.
Møller, T. (2001b). On transformations of actuarial valuation principles, Insurance:
Mathematics and Economics 28, 281–303.
Møller, T. (2001c). Risk-minimizing hedging strategies for insurance payment pro-
cesses, Finance and Stochastics 5, 419–446.
Møller, T. (2002). On valuation and risk management at the interface of insurance and
finance, British Actuarial Journal 8, 787–828.
Møller, T. (2003a). Indifference pricing of insurance contracts in a product space model,
Finance and Stochastics 7, 197–217.
Møller, T. (2003b). Indifference pricing of insurance contracts in a product space
model: applications, Insurance: Mathematics and Economics 32, 295–315.
Møller, T. (2004). Stochastic orders in dynamic reinsurance markets, Finance and
Stochastics 8, 479–499.
Musiela, M. and Rutkowski, M. (1997). Martingale Methods in Financial Modelling
(Berlin: Springer-Verlag).
Nelson, C. R. and Siegel, A. F. (1987). Parsimonious modeling of yield curves, Journal
of Business 60, 473–489.
276 References
Nielsen, J. and Sandmann, K. (1995). Equity-linked life insurance: a model with
stochastic interest rates, Insurance: Mathematics and Economics 16, 225–253.
Nielsen, J. and Sandmann, K. (1996). Uniqueness of the fair premium for equity-linked
life contracts, The Geneva Papers on Risk and Insurance Theory 21, 65–102.
Nielsen, L. T. (1999). Pricing and Hedging of Derivative Securities (New York: Oxford
University Press).
Nielsen, P. H. (2005). Optimal bonus strategies in life insurance: the Markov chain
interest case, Scandinavian Actuarial Journal, pp. 81–102.
Nielsen, P. H. (2006). Utility maximization and risk minimization in life and pension
insurance, Finance and Stochastics 10, 75–97.
Norberg, R. (1989). Experience rating in group life insurance, Scandinavian Actuarial
Journal, pp. 194–224.
Norberg, R. (1991). Reserves in life and pension insurance, Scandinavian Actuarial
Journal, pp. 3–24.
Norberg, R. (1999). A theory of bonus in life insurance, Finance and Stochastics
3, 373–390.
Norberg, R. (2000). Basic Life Insurance Mathematics. Lecture notes, Laboratory of
Actuarial Mathematics, University of Copenhagen.
Norberg, R. (2001). On bonus and bonus prognoses in life insurance, Scandinavian
Actuarial Journal 2, 126–147.
Norberg, R. and Steffensen, M. (2005). What is the time value of a stream of invest-
ments?, Journal of Applied Probability 42, 861–866.
Olivieri, A. (2001). Uncertainty in mortality projections: an actuarial perspective,
Insurance: Mathematics and Economics 29, 231–245.
Olivieri, A. and Pitacco, E. (2002). Inference about Mortality Improvements in Life
Annuity Portfolios. Proceedings of the Transactions of the 27th International
Congress of Actuaries, Cancun, Mexico.
Pham, H. (2000). On quadratic hedging in continuous time, Mathematical Methods of
Operations Research 51, 315–339.
Pliska, S. R. (1997). Introduction to Mathematical Finance: Discrete Time Models
(Oxford: Blackwell Publishers).
Ramlau-Hansen, H. (1988). The emergence of profit in life insurance, Insurance:
Mathematics and Economics 7, 225–236.
Ramlau-Hansen, H. (1991). Distribution of surplus in life insurance, ASTIN Bulletin
21, 57–71.
Rebonato, R. (2002). Modern Pricing of Interest-Rate Derivatives: The LIBOR Market
and Beyond (Princeton: Princeton University Press).
Rheinländer, T. and Schweizer, M. (1997). On L
2
-projections on a space of stochastic
integrals, Annals of Probability 25, 1810–1831.
Schönbucher, P. J. (2003). Credit Derivatives Pricing Models: Models, Pricing and
Implementation (Chichester: John Wiley & Sons Ltd).
Schwarz, H. R. (1989). Numerical Analysis. A Comprehensive Introduction (New York:
John Wiley & Sons Ltd.)
Schweizer, M. (1988). Hedging of options in a general semimartingale model. Disser-
tation, ETH Zurich, no. 8615.
Schweizer, M. (1995). Variance-optimal hedging in discrete time, Mathematics of
Operations Research 20, 1–32.
References 277
Schweizer, M. (2001a). A guided tour through quadratic hedging approaches, in
Jouini, E., Cvitani´ c, J. and Musiela, M. (eds), Option Pricing, Interest Rates and
Risk Management (Cambridge: Cambridge University Press), pp. 538–574.
Schweizer, M. (2001b). From actuarial to financial valuation principles, Insurance:
Mathematics and Economics 28, 31–47.
Sharpe, W. F. (1976). Corporate pension funding policy, Journal of Financial Eco-
nomics 3, 183–193.
Shiryaev, A. N., Kabanov, Y. M., Kramkov, D. D. and Melnikov, A. V. (1994).
Towards the theory of options of both European and American types. I. Discrete
time, Theory of Probability and its Applications 39, 14–60.
Sondermann, D. (1991). Reinsurance in arbitrage-free markets, Insurance: Mathematics
and Economics 10, 191–202.
Steffensen, M. (2000). A no arbitrage approach to Thiele’s differential equation,
Insurance: Mathematics and Economics 27, 201–214.
Steffensen, M. (2001). On valuation and control in life and pension insurance, Ph.D.
thesis, University of Copenhagen.
Steffensen, M. (2002). Intervention options in life insurance, Insurance: Mathematics
and Economics 31, 71–85.
Steffensen, M. (2004). On Merton’s problem for life insurers, ASTIN Bulletin 34, 5–25.
Steffensen, M. (2006a). Quadratic optimization of life and pension insurance payments,
ASTIN Bulletin 36, 245–267.
Steffensen, M. (2006b). Surplus-linked life insurance, Scandinavian Actuarial Journal,
pp. 1–22.
Svensson, L. E. O. (1995). Estimating forward interest rates with the extended Nelson &
Siegel method, Penning- & Valutapolitik 1995(3), 13–26.
Turner, S. H. (1969). Asset value guarantees under equity-based products, with discus-
sion, Transactions of the Society of Actuaries XXI, 459–493.
Turner, S. H. (1971). Equity-based life insurance in the United Kingdom, with discus-
sion, Transactions of the Society of Actuaries XXIII, 273–324.
Vasiˇ cek, O. (1977). An equilibrium characterization of the term structure, Journal of
Financial Economics 5, 177–188.
Wilkie, A. D. (1978). Maturity (and other) guarantees under unit linked policies,
Transactions of the Faculty of Actuaries 36, 27–41.
Index
accumulation factor, 48
additional benefits, 140
approximation price, 191
arbitrage, 23, 85, 116
attainability, 85, 118, 122, 173
basis
first order, 13
real, 15
second order, 12
binomial model, 115, 143, 175
representation formula, 176
Black’s formula
cap, 251
caplet, 251
floor, 251
floorlet, 251
swaption, 252
Black–Scholes equation, 129
Black–Scholes formula, 130
Black–Scholes model, 124,
144, 155
bonus, 222
cash, 138
terminal, 135
bonus potential, 18
collective, 19
free policy, 39
individual, 19, 29, 36, 65
premiums, 39, 43
Brownian motion, 125
cap, 239
in discrete time, 249
CMS, 240
cap, 240
in discrete time, 253
floor, 240
complete market, 118, 122, 174
constant maturity swap, see CMS
contingent claim, 117, 126
integrated, 147, 174
purely financial, 174
cost process
multi-period model, 84, 172
one-period model, 164
Cox–Ingersoll–Ross model,
95, 98
credit risk, 59
discount factor, 48
diversification, 24, 46
dividends, 207
equity-linked insurance, 148
equivalence principle, 14, 49, 206
equivalent martingale measure, 173
filtration, 79
financial standard deviation
principle, 196
financial variance principle, 196
floor, 239
in discrete time, 249
forward measure, 241
forward rate, 60
forward swap rate
in discrete time, 248
free policy, 37
hedging, 52, 118, 122
278
Index 279
incomplete market, 147, 174
integrated risk, see contingent claim,
integrated
intervention, 32, 39
investment strategy, 172
Kunita–Watanabe decomposition, 185
law of large numbers, 46
martingale measure, 87, 117, 122
mean-self-financing, 186
mean-variance hedging, 190
mean-variance indifference pricing, 194
mean-variance utility function, 194
mortality bond, 171
mortality risk
diversifiable, 50, 168
non-diversifiable, 168
systematic, 48, 170
trading with, 171
unsystematic, 48
mortality swap, 171
payer swap, 238
in discrete time, 247
payment process, 27
guaranteed, 29
not guaranteed, 29
portfolio, 121
quantile hedging, 198
receiver swap, 237
replication, 53, 118, 122
price, 174
reserve, 132
market, 18
technical, 12, 13, 104, 132, 206
total, 15, 104, 132
undistributed, 15, 105, 133
risk minimization
multi-period model, 184
one-period model, 166
risk-neutral with respect to
mortality, 162
self- financing, 84, 126, 172
strategy, 82
super-hedging, 181
price, 181
strategy, 181
surplus, 210
individual, 211
systematic, 213
surrender, 30
swap rate, 237
in discrete time, 248
swaption
in discrete time, 252
payer, 239
receiver, 239
Thiele’s differential equation
with forward rates, 64
with free policy option, 43
with surrender option, 31
unit-linked insurance, 148
indifference pricing, 197
mean-variance hedging, 193
pure, 150
risk minimization, 187
super-hedging, 182
terminal guarantee, 150
value process, 83, 116, 121, 172
variance minimization, one-period
model, 165
Vasiˇ cek model, 94, 96, 242
zero coupon bond, 50, 59

This page intentionally left blank

Market-Valuation Methods in Life and Pension Insurance
In classical life insurance mathematics, the obligations of the insurance company towards the policy holders were calculated on artificial conservative assumptions on mortality and interest rates. However, the classical approach is being superseded by developments in international accounting and solvency standards coupled with theoretical advances in the understanding of the principles and methods for a more market-based valuation of risk, i.e. its price if traded in a free market. The book describes these new approaches, and is the first to explain them in conjunction with more traditional methods. The exposition integrates methods and results from financial and insurance mathematics, and is based on the entries in a life insurance company’s market accounting scheme. With-profit insurance contracts are described in a classical actuarial model with a deterministic interest rate and no investment alternatives. The classical valuation based on conservative valuation assumptions is explained and an alternative market-valuation approach is introduced and generalized to stochastic interest rates and risky investment alternatives. The problem of incompleteness in insurance markets is addressed using a variety of methods, for example risk minimization, mean-variance hedging and utility optimization. The application of mathematical finance to unit-linked life insurance is unified with the theory of distribution of surplus in life and pension insurance. The final chapter provides an introduction to interest rate derivatives and their use in life insurance. The book will be of great interest and use to students and practitioners who need an introduction to this area, and who seek a practical yet sound guide to life insurance accounting and product development. International Series on Actuarial Science Mark Davis, Imperial College London John Hylands, Standard Life John McCutcheon, Heriot-Watt University Ragnar Norberg, London School of Economics H. Panjer, Waterloo University Andrew Wilson, Watson Wyatt The International Series on Actuarial Science, published by Cambridge University Press in conjunction with the Institute of Actuaries and the Faculty of Actuaries, will contain textbooks for students taking courses in or related to actuarial science, as well as more advanced works designed for continuing professional development or for describing and synthesising research. The series will be a vehicle for publishing books that reflect changes and developments in the curriculum, that encourage the introduction of courses on actuarial science in universities, and that show how actuarial science can be used in all areas where there is long-term financial risk.

Market-Valuation Methods in Life and Pension Insurance
THOMAS MØLLER PFA Pension, Copenhagen MOGENS STEFFENSEN Institute for Mathematical Sciences, University of Copenhagen

São Paulo Cambridge University Press The Edinburgh Building. New York. UK Published in the United States of America by Cambridge University Press.cambridge. Steffenson 2007 This publication is in copyright. or will remain. no reproduction of any part may take place without the written permission of Cambridge University Press. Møller and M. and does not guarantee that any content on such websites is.cambridge. Melbourne. Cambridge CB2 8RU. Cape Town. New York www.org/9780521868778 © T.CAMBRIDGE UNIVERSITY PRESS Cambridge. Madrid. accurate or appropriate.org Information on this title: www. Subject to statutory exception and to the provision of relevant collective licensing agreements. First published in print format 2007 ISBN-13 ISBN-10 ISBN-13 ISBN-10 978-0-511-27035-2 eBook (NetLibrary) 0-511-27035-6 eBook (NetLibrary) 978-0-521-86877-8 hardback 0-521-86877-7 hardback Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication. Singapore. .

4 A numerical example 3.6 The surrender option 2.4 The liabilities and principles for valuation 2.8 Models for the spot rate in continuous time 3.4 Dividends and bonus 1.9 Market values in insurance revisited v page ix 1 1 1 4 6 9 11 11 12 18 21 27 30 37 45 45 46 58 65 71 74 78 93 98 2 3 .2 Valuation by diversification revisited 3.1 Introduction 3.5 Unit-linked insurance and beyond Technical reserves and market values 2.3 The market-based composition of the liability 2.3 The policy holder’s account 1.2 The life insurance market 1.7 The free policy option Interest rate theory in insurance 3.1 Introduction 1.5 Bonds. interest and duration 3.6 On the estimation of forward rates 3.3 Zero coupon bonds and interest rate theory 3.5 The liability and the payments 2.7 Arbitrage-free pricing in discrete time 3.2 The traditional composition of the liability 2.1 Introduction 2.Contents Preface 1 Introduction and life insurance practice 1.

2 Swaps and beyond in continuous time 7.2 The insurance contract 6.5 Hedging integrated risk in a one-period model 5.1 Introduction 6.2 Discrete-time insurance model 4.3 The binomial model 4.4 Surplus-linked dividends 6.6 The multi-period model revisited 5.7 Surplus.7 Hedging integrated risks 5.8 Dividends linear in surplus and bonus Interest rate derivatives in insurance 7.3 The policy holder’s account 5.6 Generalizations of the models Integrated actuarial and financial valuation 5.4 Hedging integrated risks under diversification 5.6 Bonus 6.4 The Black–Scholes model 4.1 Introduction 4.and bonus-linked dividends 6.8 Traditional life insurance Surplus-linked life insurance 6. binomial and Black–Scholes 4.5 A brief introduction to market models 7.3 Surplus 6.2 Unit-linked insurance 5.5 Dividends linear in surplus 6.6 Interest rate derivatives in insurance 7.4 Swaps and beyond in discrete time 7.1 Introduction 5.3 Pricing of interest rate derivatives 7.7 A portfolio of contracts 101 101 103 115 124 131 143 146 146 148 152 161 163 172 177 199 200 200 203 210 215 217 222 226 229 235 235 236 240 246 254 257 257 5 6 7 .1 Introduction 7.5 Continuous-time insurance model 4.vi Contents 4 Bonus.

3 Risk minimization for unit-linked contracts A.1 Some results from probability theory A.Contents vii Appendix A.4 Mean-variance hedging for unit-linked contracts References Index 263 263 266 267 269 272 278 .2 The risk-minimizing strategy A.

.

being considered as two sides of the same story. Valuation and decision making on the asset side and the liability side of the insurance companies are. At the same time. These students will typically.Preface Insurance mathematics and financial mathematics have converged during the last few decades of the twentieth century and this convergence is expected to continue in the future. implementing new accounting methods to replace (assumed to be) conservative book values with real values based on market information. From this starting point. these skills need to be integrated and proven beneficial for classical problems of insurance mathematics already on a student level. and should. Demands are made on practising actuaries. International accounting standards have developed over the years. New valuation methods are added to the traditional valuation methods of insurance mathematics. By considering the convergence as it applies to their daily work. However. Many aspects of this approach are underpinned by methods taken from mathematical finance. concepts and results of finance should be brought together to construct a path between classical actuarial deterministic patterns of thinking and modern actuarial mathematics. The development has two consequences. to an increasing extent. whose education dates back to when financial mathematics was not considered as an integrated part of insurance mathematics. the ideas. such actuaries should be kept abreast of this convergence. present students of actuarial mathematics need to apply financial mathematics to classical insurance valuation problems. This is where stochastic processes are brought to the surface in payment streams as well as in investment possibilities. Denmark has been at the forefront. the Danish approach to market valuation seems to be an important step in the right direction. Although the international accounting standards have not yet been settled. meet financial mathematics in textbooks on pure finance. to receive the full benefit of financial mathematical skills. ix .

the surrender and free policy (paid-up policy) options.S. This course was organized by the Danish Actuarial Association in 2001. As a by-product. finished in 1996 (T. the with-profit insurance contract is described at first in a classical actuarial model with a deterministic interest rate and no investment alternatives. Parts of the book (Chapter 2–5) are based on material which was developed for a course on market valuation in life and pension insurance. theses. and continued our studies in our Ph.x Preface The rationale for this book is that practising actuaries need an exposition of financial methods and their applications to life insurance from the point of view of a practitioner. Various approaches to these options are suggested. and Chapter 1 was added. finished in 2000 and 2001.M. the book explains to present students how financial methods known to them can be applied to valuation problems in the life insurance market.” In 2004. and an alternative marketvaluation approach is introduced. In Chapter 2. However. written by Mogens Steffensen.D. The starting point is the version of with-profit insurance provided by Danish life insurance companies.) and 1997 (M. the partition of future payments in guaranteed payments and non-guaranteed payments is important. We studied the topics in our master theses. Tomas Björk and Ragnar Norberg gave courses in financial mathematics and applications to life insurance at the University of Copenhagen. Particular attention is paid to the intervention options held by the policy holder. Therefore. provides a non-mathematical introduction to life insurance practise in general. respectively. also written by Mogens Steffensen.e.). some definitions and introductions of quantities are repeated throughout the book. In 2002 the material was developed further and translated into English. it is still our intention that each chapter should be readable more or less independently of the others. The material was originally written in Danish. Each chapter was the material for one course module and was written more or less independently of the others. The book suggests approaches to life insurance market-valuation problems. Here. These courses aroused our interest in the interplay between finance and insurance. i. The classical valuation based on conservative valuation assumptions is explained. entitled “New Financial Products in Insurance. In 1995 and 1996. Chapters 6 and 7 were added on the occasion of The First Nordic summer school in insurance mathematics. Both discrete-time and continuous-time . In 2003. The market-valuation method introduced in Chapter 2 is generalized to a stochastic interest rate in Chapter 3. Methods and applications are discussed in terms of the Danish approach to market valuation. In order to help the reader follow the mathematical description of this type of product. the material was made consistent for notation and terminology. Chapter 1.

the cut is different and is based on entries in a life insurance company’s market accounting scheme. Fundamental financial concepts. In particular. such as arbitrage and market completeness. including the non-guaranteed payments. from mortality risk is often taken care of in the literature by assuming risk neutrality of the insurance company with respect to such a risk. Chapter 4 is written by Mogens Steffensen. for example.Preface xi bond market theory are introduced to a level such that the reader can follow the reasoning behind replacing the discount factor in the market-valuation formulas for guaranteed payments by zero coupon bond prices. Difference and differential equations for the market value of guaranteed payments are derived. we suggest various approaches to incomplete market valuation. . A stochastic stock market comes into play when valuating the non-guaranteed payments. In Chapter 4 the market-valuation method introduced in Chapter 2 is generalized to a situation with one risky investment alternative to the deterministic interest rate. An alternative class of insurance contracts to with-profit insurance is unitlinked insurance contracts. Fundamental financial concepts. are introduced in a bond market framework. Finally. Valuating the guaranteed payments. such as arbitrage and market completeness. The problem of genuine incompleteness in insurance markets is addressed. are derived. Chapter 3 is written by Thomas Møller. written by Thomas Møller.and/or continuous-time bond market model. Relaxing this assumption. The incompleteness in insurance markets stemming. the stock market is connected to the stochastic bond market introduced in Chapter 3. are repeated in a stock market framework. which are studied in Chapter 5. the important stochastic generalization of the classical actuarial deterministic financial market lies at the introduction of a stochastic bond market. This class of contracts and their market values are analyzed on the basis of the stochastic financial markets introduced in Chapters 3 and 4. In our exposition. meanvariance hedging and utility optimization are introduced and exemplified in the case of unit-linked insurance. Afterwards these are repeated in a discrete. Difference and differential equations for the market value of the total payments. Both discrete-time and continuous-time stock market theory are introduced to a level such that the reader can follow calculations of market valuations of non-guaranteed payments in the case of two investment alternatives. The usual outline of introductory financial mathematics is to introduce the fundamental financial concepts in a discrete and/or continuous stock market model. valuation and optimal investment methods based on risk minimization.

The various chapters address specific aspects of market-based valuation and contain introductions to theoretical results from financial mathematics and stochastic calculus that are necessary for the applications. The notion of surplus and various dividend and bonus schemes linked to this surplus are studied. Arbitrage Theory in Continuous Time. An introduction to certain concepts and instruments from the area of interest rate derivatives is therefore given in Chapter 7. swaptions and CMS options.xii Preface In Chapter 6. Typically. and it is demonstrated how the financial impact on a life insurance company of these instruments could be assessed. explicit results are obtained in the case where dividends and bonus payments are linear in the surplus. (2001). In contrast. In particular. Various pricing methods are discussed. whereas the financial markets typically do not offer bonds that extend more than thirty years into the future. 2nd edn (Oxford: Oxford University Press). With market-based valuation methods. written by Thomas Møller. T. The unification is based on a consideration of distribution of surplus as an integrated part of the insurance contract. the current book investigates the convergence of the theories of financial mathematics and insurance mathematics and their applications to market-based valuation. (2004). insurance companies are faced by insurance liabilities that extend up to sixty years into the future. the notation suggested by Björk is considered as “standard” and is therefore used in the present book. swaps. • Briys. Examples are swap rates. The book studies approaches for market valuation of life insurance liabilities. To some extent. Insurance: From Underwriting to Derivatives: Asset Liability Management in Insurance Companies (Chichester. The book discusses the convergence between the insurance industry and the capital markets. interest rate derivatives seem to have become an important risk-management tool for life insurance companies. It is less mathematical than the current book and focuses on institutional aspects of the interplay between the two fields. . Chapter 6 is written by Mogens Steffensen. Most of the theoretical results related to financial mathematics presented here can be found in this book by Tomas Björk. and de Varenne. E. A brief discussion of the relation to existing books on financial mathematics and insurance mathematics is given in the following list. • Björk. Even the structural exposition of certain topics of financial mathematics is inspired by Björk’s book. the application of mathematical finance to unit-linked life insurance is unified with the theory of distribution of surplus in life and pension insurance. the value of both assets and liabilities are affected by changes in the economic environment. Here. F. floors. caps. UK: Wiley).

but want to see how these disciplines can and will be combined in both theory and practice. Norberg and others). practising life insurance actuaries who need an update of the mathematics of life insurance. (1997). the current manuscript has a completely different goal and goes considerably beyond the introductory presentation in Gerber’s book. it is expected that the book will be read by students in actuarial science. We expect our readership to fall into two categories. The concepts and techniques discussed in Gerber’s book (for example. In addition. We would like to thank Tomas Björk and Ragnar Norberg for arousing our curiosity and for sharpening our understanding of the mathematics of life . specific life insurance contracts. the student also sees how aspects of the classical life insurance mathematics are implemented in practice. Firstly. However. Secondly. Basic knowledge of life insurance mathematics (such as in Gerber’s book) is required. who have prerequisites in both life insurance mathematics and mathematical finance. the theoretical results are at times developed quickly in order to get to the applications. Thiele’s differential equation) are also used and explained in the present book. The book presents a framework where the underlying insurance contracts are modeled by Markov chains and where stochastic interest rates are allowed. This provides an introduction to classical life insurance mathematics and can be viewed as a necessary prerequisite for the current book. and at some points the reader would probably benefit from studying textbooks on financial mathematics for more details and more background information.Preface xiii • Gerber. present values. the book takes the point of view of a practising actuary. Björk (2004). etc. martingales. Indeed. Chapter 1 on life insurance practice will provide the reader with sufficient insight into this practice. Life Insurance Mathematics (Berlin: Springer). for example. decrement series. The main difference between Koller’s book and the current manuscript is that we focus more on market values and the application of theories from financial mathematics in the area of life insurance. stochastic differential equations. However. Koller’s book deals more with the underlying Markov chains and on deriving differential equations for the corresponding reserves (based on the work by Hoem. By taking the viewpoint of a practising actuary as a starting point. The level is advanced.U. H. an introduction to financial mathematics in an insurance context and an approach to market valuation in life insurance. Stochastische Modelle in der Lebensversicherung (Berlin: Springer). • Koller. M. (2000). basic probability theory is required. such that the reader can follow the introduction of filtrations. No previous knowledge of financial mathematics is required.

. studies. The combination of basic knowledge in both areas and a provoked curiosity made our studies of the interplay between these fields possible.D. We would also like to thank Vibeke Thinggard and Mikkel Jarbøl for valuable comments and discussions on earlier versions of this material: in 2001 they were members of the Continued Professional Development Committee under the Danish Actuarial Association and were deeply involved in the organization of the course given in 2001.xiv Preface insurance and finance. In addition. challenging and interesting. Christian Hipp and Martin Schweizer for their guidance and support during our Ph. we wish to thank Ragnar Norberg.

exchanges a stream of payments with another party. Defined contributions with partly defined benefits cover the majority of life insurance policies. We take as our starting point the idea of the policy holder’s account. and this chapter explains in words the construction and the elements of this equation and its role in accounting. a 1 .2 The life insurance market In this section we explain the most typical environments for negotiation and contractual formulations for a life insurance policy. In life insurance. however. 1. for example. The exchanged streams of payments form. the insurance company. We distinguish between defined benefits and what we choose to call defined contributions with partly defined benefits. The purpose is to give the reader sufficient insight to benefit from the remaining chapters. The policy holder agrees with the insurance company about a certain premium to cover a basket of benefits with. in a sense. we think of the way this exchange of payments is handled and settled by the insurer. the basis of the insurance contract and the corresponding legal obligations.1 Introduction and life insurance practice 1. Note.1 Introduction This chapter provides an introduction to life insurance practice with focus on with-profit life insurance. the policy holder. This account can be interpreted as the policy holder’s reserve in the insurance company and accumulates on the basis of the so-called Thiele’s differential equation. that the policy holder’s account is not in general a capital right held by the insured but a key quantity in the insurer’s handling of his obligations. Its formulation as a forward differential equation plays a crucial role. one party. When speaking of life insurance practice.

The policies with defined contributions with partly defined benefits are naturally classified as private. These organizations typically take care of people employed in the uniformed services or education.2 Introduction and life insurance practice life insurance sum paid out upon death before the termination of the contract and/or a pension sum paid out upon survival until the termination of the contract. or optional. and this surplus is to be paid back to the policy holder. As the market interest rates. Although the terms of the contract are typically negotiated between the employer and the company. The result is an agreement where the employer is . This typically happens by increasing one or more benefits. A private policy is agreed upon by a private person and the company. The private person is the policy holder and negotiates the conditions in the contract. Since the employer has no claims and no obligations besides paying the premium. firm-based. A labor-based policy works in many respects as a firm-based policy except for the fact that the employer and the employees are represented at the negotiation by organizations rather than the employers and employees themselves. insurance risk and expenses. This combination of known premiums reflected in guaranteed benefits. in which case it is up to the employees to decide whether or not to participate. categorizes the contracts as defined contributions with partly defined benefits. the market insurance risk and the market costs evolve. the employees are still the policy holders. The total premium paid is typically a percentage of the salary. a surplus emerges. Such rare constructions should in principle be categorized as defined benefits with partly defined contributions since then a basket of known benefits is combined with guaranteed premiums which may be decreased depending on the development in the market. this premium can in many respects be interpreted as salary. which may be increased depending on the development in the market. Only very rarely is the surplus paid out in cash or used to decrease the premium instead of increasing the benefits. One construction is to increase all benefits proportionally such that the ratio between. for example. the death sum and the pension sum is maintained. in which case all employees are forced to participate. The benefits agreed upon at issuance are set systematically low by basing them on conservative assumptions on interest rates. The agreement between the employer and the company may either be compulsory. Another construction is to keep the death sum constant or regulated with some price index while residually increasing the pension sum. or labor-based. The employer typically pays a part of the premium but receives no benefits. A firm-based policy is a contract which is part of an agreement between an employer and an insurance company.

all examples and interpretations take the defined contributions with partly defined benefits policy as a starting point. The premium part paid by the employer can in many respects be interpreted as salary.2 The life insurance market 3 obliged to participate in the employees’ policy on terms agreed upon by the organizations. instead of sharing the premium defined as a percentage of the salary between the employer and the employees. only the employees’ part of the premium is defined as a percentage of the salary. a certain part of the defined contribution policies. in Denmark the national pension scheme is a pay-as-you-go scheme where present retirement pensioners are covered by present tax payers. the insurance company is pure administrator and takes no risk. In contrast. then neither the employer nor the policy holder participates in the development in the market but leaves all risk to the insurance company. As mentioned above. as:eksist It should be mentioned that in addition to the life insurance market described above. there may be a lot of differences in the concrete formulations of the various agreements and contracts. are therefore and comparable with the firm-linked policy described above. The defined benefit policies are usually a part of an agreement between a firm and an insurance company. if the risk is left to the employer exclusively. In the following. Policies belonging to different classes may also be mixed within an agreement and within a contract. When discussing details. In addition. The total premium paid is typically a percentage of the salary. it works basically as a firm-based policy with the employer and the employees as payers of premiums and the employees as receivers of benefits. The policy is then issued by a company taking care of all the employees in a particular organization. both the employer and the employed are obliged for which to agree to minimum conditions. This leaves a risk on the premium side. For instance. On the other hand. where the surplus is redistributed as cash or used to decrease the premiums. The classification given above is fairly broad. The contract is negotiated indirectly by settlement of the agreement. If the risk is left to the insurance company exclusively. The labor-based policy is often part of a compulsory agreement. there may exist a set of public insurance schemes. the benefits are also defined as a percentage of the salary. The employees are the policy holders. This risk is split between the employer and the insurance company according to the agreement. However. can actually be considered as defined benefit policies with partly defined contributions. we concentrate on defined contributions with partly defined benefits. the Danish state regulates a couple of particular funded pension schemes for people who work. . Although most of the ideas presented in Chapters 2–5 may be applied to defined benefits as well.1. Once the agreement has been made and the contract has been issued.

the outgo on an insurance account consists of the benefits paid to the policy holder. The technical reserve can in certain respects be interpreted as a bank account. The death sum exceeds the technical reserve such that the risk premium is positive and can be considered as an outgo. The risk premium is a premium that the policy holder pays from their account. the income on an insurance account consists of premiums paid by the policy holder and return provided by the insurance company from capital gains on investment of the account. Correspondingly. which can be considered as an income or an outgo depending on its sign. The other way classifies the changes in what was agreed beforehand in a particular sense and additional changes made by the insurance company at the discretion of the company. the policy holder’s account or the technical reserve and its dynamics are the technical elements in the handling and administration of life insurance contract. The outgo on a bank account consists of capital withdrawals. in which case it can be considered as an income. two additional terms add to the change of the account.4 Introduction and life insurance practice 1. the loss to the insurance company in case of death equals the death sum which has to be paid out minus the technical reserve which. And.3 The policy holder’s account In our setting. the insurance company cashes the technical reserve upon death and has no obligations. The first way classifies the changes of income and outgo. One term is an outgo and covers the expense to the insurance company to administrate the policy. This results in a negative risk premium. but these are charged indirectly by a reduction of the return. and the . in which case it can be considered as an outgo. Administration expenses on the bank account must also be paid. However. it may be positive. The expected value of that loss is the difference between the death sum and the technical reserve times the probability of dying in some small time interval. can be cashed. The income on a bank account consists of capital injections and capital return provided by the bank from capital gains on investment of the account. The expected value of this gain is the technical reserve times the probability of dying in a small interval. The other additional term which adds to the change of the account is the so-called risk premium. The premium for this coverage is set to the expected value of this loss. correspondingly. Considering a so-called term insurance paying out a sum upon death. We classify the changes of the technical reserve in two different ways. The risk premium is paid to cover the loss to the insurance company in case an insurance event takes place in some small time interval. or negative. on the other hand. Considering instead a pure endowment insurance. The amount of the potential loss is also spoken of as the sum at risk.

Concerning the mortality. The usual way of allocating the dividends is to change the account. Secondly. but the underlying idea is basically the same. it guarantees a minimum benefit which is based on a technical return. the technical change and the additional change. it pays back the difference between the risk premiums in correspondence with the more favorable probability and the technical probability. Another way of classifying the changes of the technical reserve is firstly to identify the technical change which conforms with the guaranteed payments and then identify the additional changes made by the insurance company at the discretion of the company. returns and benefits. it is based on a technical amount for administration expenses. not in correspondence with the technical assumptions. the picture becomes more blurred. insurance risk and expenses. it pays back the difference between the more favorable amount and the technical amount for expenses. premium waiver and deferred benefits. Concerning the expenses. This surplus is provided by the policy holder due to conservatism in assumptions and has to be paid back as the real market conditions evolve. the insurance company firstly withdraws the technical amount for expenses. it pays the difference between the more favorable return and the technical return. it simply guarantees to pay out a benefit which is “fair” under a certain set of assumptions on return. Then. this set of assumptions is meant to be set so conservatively that a surplus emerges over time. the policy holder . the policy holder pays or gains. Apart from the real incomes and outgoes in form of premiums. the bank account can be interpreted as an insurance contract where the technical reserve is simply paid out upon death. element by element. The law states that the surplus must be paid back to those who created it.3 The policy holder’s account 5 risk premium may be considered as an income. The term insurance and the pure endowment insurance are simple insurance contracts. However. This gives a potential loss upon death of the technical reserve paid out minus the technical reserve cashed. depending on the sign of the risk premium. The use of the favorable assumption is that. for the risk imposed on the insurance company. When an insurance company issues an insurance policy. for example the probability of dying in a small time interval. Basically. Finally. Furthermore. the minimum benefit is based on a certain technical probability of the insurance event. Secondly. Concerning the return. but in correspondence with a set of assumptions that is more favorable to the policy holder. This is achieved by adding dividends to the policy holder’s account.1. the insurance company firstly collects a risk premium in correspondence with the technical probability of death. element by element. Secondly. In addition. If we introduce such things as disability annuities. whereby the risk premium equals zero. we can classify. the insurance company firstly pays the technical return.

We should mention an alternative application of dividends which has gained ground in recent years. In practice. the consequence of less favorable technical conditions is higher surplus contributions in the future. secondly. Instead of increasing benefits and/or decreasing premiums. and. and it is stated by law that this surplus should be paid back to those policy holders who created it. allocation of dividends in a way postpone the increment of guaranteed benefits without postponing the increment of the account. .4 Dividends and bonus The premiums agreed upon at the time of issuance of an insurance policy are “too high” compared with the benefits that are guaranteed at the time of issuance. Furthermore. benefits and risk premiums need to be adjusted in the light of such a change of payments. One may then ask: where did the money go and does allocation of dividends really put the policy holder in a better position? The point is that paying out dividends leads to what seems to be less favorable technical conditions. The favorable development of the account including dividends may be used to reach a higher value to be paid out as a pension sum at the termination of the contract. such that the ratio between the two benefits is constant. since premiums. And since these surplus contributions eventually have to be redistributed and reflected in payments. the surplus is distributed among the owners of the insurance company and the group of policy holders. this happens in two steps.6 Introduction and life insurance practice should not be put into a worse position than if the technical assumption had been used. However. dividends are added to the policy holder’s account without changing the guaranteed payments. one may keep the guaranteed payments and instead change the underlying technical assumptions. 1. the policy holder may also wish that this favorable development provides capital for an increase of benefits and/or a decrease of premiums throughout the term of the policy. This disproportion is the source of surplus. However. One construction is to let the death sum and the pension sum increase proportionally. the most typical constructions are to let the death sum be constant or regulated by some price index and then use the residual dividends to increase the pension sum. the guaranteed payments are not changed and can therefore not be less favorable. the surplus distributed to the policy holders is distributed among the policy holders. the position is indeed favorable. In this way. This has a feedback effect on the dynamics of the account. However. Firstly. By this construction.

The insurance company is allowed to invest not only in fixed income assets. However. the distribution of surplus between owners and policy holders has to be fair in some sense. are fair. closely connected to the risk of the insurance company owners to eventually suffer a loss on the portfolio. a risky business. This would systematically redistribute surplus from the past and present policy holders to the future policy holders. when things go right. A redistribution of the surplus to those who earned it has two consequences. Fairness is here given by the statement that the surplus should be redistributed to those who earned it. of course. Thus. the insurance company needs to assign the undistributed reserve to the individual technical reserves “in due time. to an increasing extent. and the insurance company may end up in a situation where it is not possible to increase the policy holder’s account by the technical interest rate by means of capital gains. The risk that things may go wrong. Also. Investment in stocks is. is the reason why they. the mutual distribution between policy holders is also required to be fair.4 Dividends and bonus 7 So. The distribution among policy holders takes place by deciding on the favorable set of assumptions introduced above. That is. The part of the surplus distributed to the policy holders is deposited in the so-called undistributed reserve. sociological and demographic uncertainties play a different role. why should the owners of the company take part in the surplus that was created by policy holders? The problem is that “too high” may not be high enough. leading to the owners having to pay. however. The owners of the insurance company must eventually cover the loss on the insurance portfolio. “due time” is. In that situation the owners of the insurance company must still provide capital for the technical increments of the technical reserve. this reserve is distributed to the policy holders as a group but is not yet distributed among the policy holders. This mechanism transfers money from the undistributed reserve to the individual policy holder’s account. medical. Concerning mortality and other kinds of insurance risk. The first consequence is that the insurance company is not allowed to grow “large” undistributed reserves. One of the purposes of this book is to provide the insurance companies with tools and ideas to make distributions that. The insurance company may need to help by injecting capital in the technical reserve in order to live up to the technical conditions. which again connects to the owner’s share in . but also in stocks.” Here.1. deserve a share in the surplus created by the insurance portfolio. concerning mortality and expenses the insurance company may experience a situation worse than that considered as the worst possible case at the time of issuance. As was required from the distribution between owners and policy holders.

If the undistributed reserve is low. make up the conditions for all changes that are made over time by the insurance company to the premiums and benefits agreed upon at issuance. in the case of a firm-linked or labor-linked contract. in combination. the set of favorable assumptions must to some extent reflect the present policy holder’s contributions to the undistributed reserves. premium to the owners for taking risk.8 Introduction and life insurance practice the surplus. in many respects. the risk premium is proportional to the sum at risk. the conversion of dividends into payments on the individual policy is a part of the individual policy conditions. the sum at risk and the premium determine the individual share in the total distribution. and the expense is typically formalized as a part of the premium. The redistribution of the surplus between the owners of the company and the policy holders and the mutual redistribution between policy holders are regulated by law and overseen by the supervisory authorities. Therefore. Therefore the individual technical reserve. Thus. then this reserve can take a big loss before the owners have to take over. and the insurance portfolio must pay a larger premium to the owners. mortality and expenses. Depending on the bonus scheme. The return is proportional to the technical reserve. Firstly. for example. . the distribution between owners and policy holders (regulated by law). Such a mechanism can be imposed by favorable assumptions on interest rates. then this reserve may easily run out. after a price index regulation of the death sum. the policy holder may experience the redistribution in different ways. This shows that the solution to a fair distribution of the surplus between the owners and the policy holders interacts substantially with the solution to a fair mutual redistribution amongst policy holders over time. The typical construction is to increase the benefits proportionally or to increase the pension sum residually. It is important to realize how the legislative environment and the contract. If the undistributed reserve is high. as part of the contract itself. this conversion is directly negotiable between the insurance company and the policy holder or. The second consequence is that. On the other hand. Then the insurance portfolio pays a small. they are not directly specified in the contract. However. they make up a part of the legislative environment in which the contract has been agreed upon. The redistribution may also be paid out as cash. this must happen in a way that reflects which present policy holders have contributed a lot to the surplus and which have contributed less. respectively. given a redistribution to the present policy holders. between the company and the firm or labor organization. and therefore they can be considered. possibly zero. Once a redistribution from the undistributed reserve among the policy holders is elected to happen now instead of later.

no problem in maintaining the smoothing effect in a unit-linked insurance policy. together with the other policy holders and together with the owners of the insurance company. One way of avoiding the problems with fair distribution is that each and every single policy holder forms their own individual fund. but since the beginning of the twenty-first century life insurance companies in these countries have started to offer unit-linked insurance contracts. Some insurance companies have introduced advanced . policy holders typically also give up certain features of the participating policy. In a unit-linked insurance contract. so to speak. this may be a difficult task. This is what happens in unit-linked insurance. 1. The unit-linked insurance contract can be decorated with many different kinds of guarantees. and hereby the guaranteed payments. it may be difficult for the individual policy holder to understand whether the conditions are followed. Even by representation of their ambassadors in the cooperation in the form of the supervisory authorities. When giving up the investment cooperation and entering into unit-linked contracts. the market is still young. However. from shocks in the market conditions. one achieves a smoothing effect of the market conditions. The policy holder participates in a mutual fund. The participating policies hold a very strong position in many countries and the unit-linked market has been long in coming. The technical reserves then only experience a smoothed effect from such shocks. However. it is important to realize that these smoothing effects do not rely particularly on the policy holders’ participation in an investment cooperation.5 Unit-linked insurance and beyond Several features characterize the participating policy as explained above. the distribution mutually among policy holders (regulated by law). in principle. and insurance companies have shown some creativity on that point. The legislative environment sets the conditions for this cooperation. the policy holder does not participate in a mutual fund but decides on their own investments to some extent. by changing the terms in the individual policy (regulated by the contract).1. and thirdly. By working with an undistributed reserve. There is. This is only a matter of a proper definition of the unit to which the payments of the contract are linked. The undistributed reserve protects the underlying technical reserves.5 Unit-linked insurance and beyond 9 secondly. in particular concerning the several layers of fair distributions. and there is still a lot of space for new developments and improvements. However.

a proper description of unit-linked products in terms of finance theory requires an enlargement of this environment. One may argue that a unit-linked insurance contract endowed appropriately with smoothing effects and guarantees is close. However. both in spirit and in payments. One challenge is to incorporate the participating policies in an environment of finance theory. one may also argue that it makes a huge difference whether the conditions for smoothing effects and the guarantees are stated in the contract and individualized or are given in the legislative environment by somewhat more vague statements on fairness. If the investment game is individualized. our characterization of unit-linked products is that the investment game is an individual matter. On the other hand. as has successfully been achieved for unit-linked policies. One of the aims of the remaining chapters of this book is to provide the reader with a box of tools that can be applied for working with this challenge with theoretical substantiation. no matter the complexity of the unit. . Furthermore. even when including any kind of smoothing effect. then the unit-linked contract stays unit-linked.10 Introduction and life insurance practice unit-linked products which maintain the smoothing effect. an appropriate enlargement of this environment is definitely needed to deal with the complex nature of participating contracts and the special conditions of the life and pension insurance market in general. When speaking of such products as unit-linked contracts. to a participating policy.

This primary example is an endowment insurance with premium intensity . The purpose of the chapter is to demonstrate the retrospective accumulation of the technical reserve and to formalize an approach to prospective market valuation.1 Introduction This chapter deals with some aspects of valuation in life and pension insurance that are relevant for accounting at market values. taking into consideration more realistic actuarial and financial modeling. The idea of a liability as a retrospectively calculated quantity needs revision when going from the traditional composition of the liability to a market-based composition of the liability. see. Generalizations to other insurance contracts are left to the reader. bad . Throughout the chapter expenses are disregarded. things are turned somewhat upside down here. secondly the prospective approach to market valuation and thirdly the underpinning principles. By considering firstly the retrospective accumulation of technical reserves. and guaranteed death sum. This is an important step towards comprehending both the market-valuation approach presented here and the generalization and improvement hereof. The aim is to meet the practical reader at a starting point with which he is familiar. pension sum guaranteed at time 0. for example. we consider all calculations pertaining to the primary example of an insurance contract. 11 . Norberg (2000) or Steffensen (2001). The exposition of the material distinguishes itself from scientific expositions of the same subject. The terms prospective and retrospective play an important role. ba 0 .2 Technical reserves and market values 2. We explain and discuss the principles underpinning this approach. The insurance contract is issued at time 0 when the insured is x years old and with a term of n years. Throughout the chapter. Bonus is paid out by increasing the pension sum.

1 The technical reserve and the second order basis In this section we study the technical reserve and how it arises from the second order basis. Another important quantity is the survival probability from time t to u corresponding to age x + t u u to age x + u. on the other hand. and assume that discounting is based on a deterministic interest rate r. x+t but since only continuous-time versions appear in this chapter. keeping in mind that this s of course depends on x then.1) . We consider an accumulation of the technical reserve by the second order basis. which we abbreviate by exp − t . for example n−t Ex+t and ax+t n−t .2. 2. exp − t s ds . with the decorations ∗ ∗ and . the quantities ax+t n−t and A1 n−t should be decorated with a bar. a temporary term insurance and a temporary life annuity also used as a premium payment annuity: n−t Ex+t = vn−t n−t px+t = e− n t n t n t r+ A1 x+tn−t = ax+tn−t = vs−t s−t px+t s ds = n t n t s t e− r+ s t r+ s ds vs−t s−t px+t ds = e− ds In principle. These refer to the corresponding fundamental quantities r and and the contents are obvious from the context.2 The traditional composition of the liability 2. where v = exp −r . The actuarial notation for such a survival probability is u−t px+t . The second order basis is a pair r containing the second order interest rate and the second order mortality rate by which the technical reserve accumulates.12 Technical reserves and market values Assume that death occurs with a deterministic mortality intensity s at age x + s. All quantities appear. If the interest rate is constant. This corresponds to a difference equation with initial condition as follows: V∗ t = r V∗ 0 = 0 t V∗ t t+ t− t R∗ t t (2. this notation is omitted. which u we abbreviate by exp − t r . We remind the reader about the following notation for the capital value at time t of one unit of a pure endowment insurance. exp − t r s ds . this discount factor equals exp −r u − t = vu−t . An imporu tant quantity is the discount factor from u ≥ t to t.

The corresponding differential equation with initial condition reads as follows: d ∗ V t =r dt V∗ 0 = 0 This differential equation can be solved over the time interval 0 t and the initial condition leads to the retrospective form: V∗ t = t t V∗ t + − t R∗ t (2.1) is a discrete-time version of a corresponding differential equation which can be obtained by dividing the difference equation by t and letting t go to zero. if the terminal value V ∗ n is interpreted as a terminal benefit.2 The technical reserve and the first order basis In this section we study the technical reserve and how it arises from the first order basis. Nevertheless. solved over the time interval t n leads to the prospective form.2. The second order basis is a decision variable held by the insurer that is to be chosen within certain legislative constraints and market conditions. One notes that in the prospective form. Equation (2. Equation (2.4) where V ∗ n is the terminal value of the technical reserve. . The first order basis is a pair r ∗ ∗ containing the first order interest rate and the first order mortality rate under which the guaranteed benefits are set according to the equivalence principle.4).2. Equation (2. 2.3) The differential equation. Since these may be unknown at time t. V ∗ t = bad A1 n−t + V ∗ n x+t n−t Ex+t − ax+tn−t (2.4) is a representation of the solution to Equation (2.2). but not a constructive tool for calculation of the technical reserve.2) e 0 t s r + − bad s ds (2. Equation (2. the second order basis over t n appears together with the terminal value of the technical reserve V ∗ n . the prospective form expresses the technical reserve as a prospective value of all payments valuated under the future second order basis.2).2 Traditional composition of the liability where R∗ is the sum at risk given by R∗ t = bad − V ∗ t 13 and where t is the time unit chosen for the accumulation. Actually.

any pair r Equation (2. In the following. Note that Equation (2.9) Note that from Equation (2.7) On the basis of the technical reserve. we see that Equation (2.5).5) t − r∗ t V ∗ t + t − t R∗ t (2.8) we get ba n = V ∗ n such that the technical reserve at the terminal time n coincides with the pension sum at that time.6) is a candidate for the second order basis. the prospective . we write ba n instead of V ∗ n when it is appropriate to think of V ∗ n as the pension sum. Nevertheless. The differential equation. the pension sum guaranteed at time t is calculated in accordance with the equivalence principle. we can write the prospective form as follows: V ∗ t = bad A1∗ n−t + ba t x+t ∗ n−t Ex+t − a∗ n−t x+t (2. Equation (2.1 coincides with the technical reserve in this section.14 Technical reserves and market values We consider an accumulation of the technical reserve by the first order basis. This motivates the interpretation of the technical reserve V ∗ n as the terminal benefit at the end of the Section 2. given a rate of dividends . This corresponds to a differential equation with initial condition as follows: d ∗ V t = r∗ t V ∗ t + dt V∗ 0 = 0 where the dividend rate t = r is given by ∗ − ∗ t R∗ t + t (2.2. solved over 0 t leads to the retrospective form.8) also sets the guaranteed pension sum at time 0: ba 0 = a∗ n − bad A1∗ x xn ∗ n Ex Since ba t is calculated on the basis of the retrospectively derived technical reserve. On the conforming with other hand. ba t = V ∗ t + a∗ n−t − bad A1∗ n−t x+t x+t ∗ n−t Ex+t (2.2.6). V∗ t = t e 0 t ∗ ∗ s r + − bad ∗ s + s ds (2.6) The rate of dividends is determined such that the technical reserve in Section 2.1.8) Hereafter. Given a second order basis the rate of dividends is determined by Equation (2.9) is a representation of V ∗ t but not a constructive tool for its derivation.

we use a technical reserve at time u ≥ t for payments guaranteed at time t.2 Traditional composition of the liability 15 form expresses the technical reserve as a prospective value of the payments guaranteed at time t valuated under the first order basis.12) . given by R t = bad − U t The differential equation.2. Equation (2.10) 2. This corresponds to the differential equation with initial condition as follows: d U t =r t U t + dt U 0 =0 where R is the sum at risk. given by R∗ t u = bad − V ∗ t u − ∗ u R∗ t u (2. respectively. solved over 0 t leads to the retrospective form: U t = t − t R t (2. Then a V ∗ t u = bad A1∗ x+un−u + b t ∗ n−u Ex+u − a∗ x+un−u and for u ≥ t we have a differential equation with initial and terminal conditions.11) e 0 t s r+ − bad s ds (2. as follows: u V ∗ t u = r∗ u V ∗ t u + V∗ t t = V∗ t V ∗ t n = ba t where R∗ t u is the sum at risk.3 The undistributed reserve and the real basis In this section we study the undistributed reserve and discuss how the undistributed reserve arises from the real basis.11). The technical reserve at time t covers payments guaranteed at time t. The real or third order basis is the pair r containing the real or third order interest rate and the real or third order mortality rate by which the total reserve accumulates. Later. We choose to introduce this quantity now and denote it by V ∗ t u .2. We consider an accumulation of the total reserve by the real basis.

20) holds.18) shows how the undistributed reserve is consumed in the future by letting the dividend rate differ from the contribution rate.14) e 0 t s r+ c s − s ds (2.19) the condition on the second order basis. with V t defined as the prospective value of future payments as follows: V t ≡ bad A1 n−t + ba n x+t n−t Ex+t − ax+t n−t (2.20). solved over t n leads to the prospective form: U t = bad A1 n−t + U n x+t n−t Ex+t − ax+t n−t (2. we work under the assumption from Equation (2. Now we put up the following condition on the second order basis: X n = U n −V∗ n = 0 (2. Equation (2.17) Under this condition we have the following prospective form for the undistributed reserve: X t = n t e− s t r+ s − c s ds (2.18) with respect to t.16 Technical reserves and market values The differential equation.15) and (2. Equation (2.11). .13) where U n is the terminal value of the total reserve. We therefore switch between U and V . implies that U t =V t (2.18) This can be verified by differentiation of Equations (2.13) by V ∗ n = ba n . Equation (2.17).20) This is obtained by replacing U n in Equation (2. from situation to situation depending on whether it is beneficial to think of the quantity as retrospective or prospective.17) such that Equation (2. in accordance with Equation (2. we obtain the retrospective form: X t = t (2. In this chapter.15) where the contribution rate c is defined by c t = r t − r∗ t V ∗ t + ∗ t − t R∗ t (2.16) The retrospective form given in Equation (2. The undistributed reserve X is calculated residually as the difference between the total reserve and the technical reserve: X t = U t −V∗ t By differentiation.15) shows how the undistributed reserve consists of past contributions minus past dividends. Furthermore.

12) and (2. following from the condition X n = 0. The total reserve is a value of the total payments and the undistributed reserve is a value of dividends minus contributions. of the total and the undistributed reserves but are not constructive tools for their derivation. for one reason or another.2 Traditional composition of the liability 17 Note that the future second order basis appears in Equations (2. the formulas above show that further specification of the future second order basis is redundant.18) and (2.15).12) and (2. It is important to understand that it is the condition X n = 0 that spares us from discussing the future second order basis further. one may also start out with these representations as definitions. respectively. in the traditional composition of the liability. the sum at risk and the premium. together with other sources of capital. The purpose of Equation (2. Equations (2. one would. However. one would not accept that U ≤ V ∗ . n The condition U n − ba n = 0 can. This means that Equations (2. One of the potentials of accounting at market value is to set up alternative solvency rules in terms of market values. including the safety margins this may contain. the condition X is not to be fulfilled. If.19) through dividends and the terminal pension sum. excess a value based on the technical reserve.21) is to secure that U at any point in time covers the technical reserve interpreted as the value of guaranteed payments valuated by the first order basis. be written as follows: ba n n Ex + bad A1 − axn = 0 xn This shows that the condition X n = 0 is the same as performing the equivalence principle on the total payments under the real basis. impose a solvency condition in the following form: X t ≥0 0≤t≤n (2.20) are representations of the same quantities based on the condition X n = 0. Equations (2.15) are definitions of U and X whereas Equations (2. If we then restrict the second order basis to obey X n = 0 with X defined by Equation (2. in addition to the terminal condition X n = 0. by multiplication of e− 0 r+ . As we shall .18) and (2.21) In practice.15). hereby including the future second order basis in the definition of U and X. the future second order basis is inevitably brought into the quantity V .2.18) and (2. these forms express the total and the undistributed reserves as prospective values of different future payments depending on the future second order basis valuated under the real basis. Nevertheless. The quantities are hereafter given simply by the retrospective formulas. Thus. the undistributed reserve must.19) are representations. Traditionally.

We choose to introduce this quantity now and denote it by V g t u . The guaranteed payments at time t are given by the premium rate .3. Later. Also in this section we define prospective values based on the . we use a market reserve at time u ≥ t for the payments guaranteed at time t.22) Thus. as follows: d g V t u = r u Vg t u + du Vg t t = Vg t V g t n = ba t where Rg t u is the sum at risk.3 we introduced U and X and we concluded that the retrospective quantities were constructive tools for the calculation of the prospective quantities given by Equations (2. Then V g t u = bad A1 n−u + ba t x+u n−u Ex+u − ax+u n−u and for u ≥ t we have a differential equation with initial and terminal conditions.18) and (2. in principle.1 Guaranteed payments and the market reserve In this section we introduce the market reserve for the guaranteed payments. given by Rg t u = bad − V g t u − u Rg t u (2. V g t .20).18 Technical reserves and market values see.2.23) 2. respectively.3. solvency rules such as U ≥ V ∗ are.2 Bonus payments and the bonus potential In this section we introduce the bonus payments.3 The market-based composition of the liability 2. the bonus potential. V g t is a prospective value of the guaranteed payments under the real basis. given the condition on the second order basis. not necessary in the market-based composition of the liabilities. 2. The market reserve at time t is given by the prospective formula. The market reserve at time t. covers payments guaranteed at time t. the death sum bad and the pension sum ba t . V g t = bad A1 n−t + ba t x+t n−t Ex+t − ax+t n−t (2. X n = 0. In Section 2. the individual bonus potential and the collective bonus potential.

as follows: u Vb t u = r u + Vb t t = Vb t V b t n = ba n − ba t Now.24) depends on the future second order basis through the terminal pension sum. The real payments differ from these payments by the pension sum ba n only. we use a market-based reserve at time u ≥ t for the payments not guaranteed at time t. Later. it can now be shown that V b t u = U u − V g t u .25) We note that the bonus potential at time 0 equals the negative market reserve.15). We choose to introduce this quantity now and denote this by V b t u . As we shall see. By means of the differential equations (2. covers payments not guaranteed at time t.11) and (2.3.3 Market-based composition of the liability 19 future payments. the death sum bad and the pension sum ba t . we include the condition on the second order basis that X n = 0 with X given by Equation (2. the quantity V b defined by Equation (2. with X given by Equation (2. −V g 0 .26). the condition X n = 0. as in Section 2. again leads to retrospective calculation formulas.26) u Vb t u (2. . Then V b t u = ba n − ba t n−u Ex+u and for u ≥ t we have a differential equation with initial and terminal conditions. V t is a prospective value under the real basis of the payments not guaranteed. In general.24) Thus. It turns out to be informative to decompose the bonus potential into two reserves: the individual bonus potential V ib and the collective bonus potential V cb . We also speak of V b t as the bonus potential at time t. The guaranteed payments at time t are given by the premium .2. from which it follows that Vb t = U t −Vg t (2. This bonus payment has the following market value: V b t = ba n − ba t b n−t Ex+t (2. However. The difference is exactly the pension sum ba n − ba t making up the bonus payments. the condition X n = 0 saves us from further specifications since we have the representation in Equation (2. respectively. The bonus payments at time t are the payments which are not guaranteed at time t.15).2. since U 0 = 0.23). V b t . The bonus potential at time t.

From the differential equations (2.29) Thus. we discuss alternative definitions of the individual bonus potential.23) and the terminal condition V ∗ n t − V g n t = ba t − ba t = 0. Secondly.3.27) does not necessarily hold. since V ∗ 0 = 0. V ib t = V ∗ t − V g t = where c t s = r s − r∗ s V ∗ t s + ∗ n t (2.32) In Sections 2. we consider the collective bonus potential.5. one obtains the prospective form.10) and (2. consider the individual bonus potential. If V ∗ < V g . We note that the collective bonus potential at time 0 equals zero since U 0 = V ∗ 0 = 0. Taking into . we set the individual bonus potential to zero using the following formula: V ib = V ∗ − V g + + = max V ∗ V g − V g where a equals a if a ≥ 0 and 0 if a < 0. This is equivalent to V ib t = n t + r+ (2.1 and 2. We note that the individual bonus potential at time 0 equals the negative market reserve.27) e− s t r+ c t s ds (2. This explains the term individual bonus potential.20 Technical reserves and market values We consider the situation V ≥ V∗ ≥ Vg Then V b is decomposed as follows: V b = V cb + V ib = V − V ∗ + V ∗ − V g Firstly. the individual bonus potential is simply the market value of the safety margins in the guaranteed payments.31) e− s t c t s ds (2. −V g 0 .28) s − s R∗ t s (2.6. We now turn to the general case where Equation (2. V cb t = V t − V ∗ t = X t (2. it may also happen that V < V ∗ . In that case we set the collective bonus potential to zero.30) The collective bonus potential can also be calculated residually as the bonus potential minus the individual bonus potential. This explains the term collective bonus potential. If we do not have the solvency constraint X t ≥ 0.

We now discuss the elements in these prospective formulas.12). (2. . V ∗ and V g . is called an indicator process for the condition which has to be fulfilled for the process to assume the value 1. and in Section 2.24).4 Liabilities and principles for valuation 21 account all possible relations between V .22). 2. Z is an indicator process for the insured to be dead. assuming the values 0 or 1. (2.4). We represent these liabilities as conditional expected values and discuss the principles underpinning this representation. A process is here defined as a continuum of stochastic variables. These benefits appear as building blocks in a number of insurance types. Thus. N t = number of deaths until time t. This also helps us in the search for theoretically substantiated methods for calculation of values in the case where we do not require from the second order basis that X n = 0. such that Z at time t assumes a certain value Z t . and in this situation. which is the main example given in this chapter.9).18). (2. We introduce another process N counting the number of deaths.2 we presented formulas of both retrospective (Equations (2.7). Assume that Z is a process measuring whether death has occurred.e. This enables us to generalize these formulas to other types of insurance. a temporary term insurance or a temporary life annuity. We remind the reader that a liability is a value set aside by the insurer in order to be able to meet certain obligations in the future. In Section 2.4 The liabilities and principles for valuation In this section we interpret the liabilities put up in the previous sections. indexed by time. (2.3).28). Elementary examples of such payments are the benefits of a pure endowment insurance.33) Such a process. (2.30)). (2. Then Z t = 0 1 if death has not occurred at time t if death has occurred at time t (2.20)) type.15)) and prospective (Equations (2. of course. for example the level premium paid endowment insurance.3 we presented primarily prospective formulas (Equations (2. (2. (2. (2. Z = N . i.2. the collective and the individual bonus potential are formalized by the following general formulas: V cb = V − V g − V ∗ − V g + + + + V ib = V − V g − V − V g − V ∗ − V g From these formulas we easily see that V cb and V ib are positive and sum up V b .

1 Ax+t n−t .1.e.35) On the left hand side of Equation (2. If the insured survives until time n. Whereas the elementary payments are just what we want to s . The process I is defined by I t = 1 − Z t = 1 − N t . The quantity dN s is the change in N at time s. conditional expectation Et0 · and the elementary payments I n I s ds and dN s ). The integral now becomes a sum of infinitely many zeros plus one discount factor from Tx to t if death occurs before time n. We now interpret the elements in the square brackets in Equations (2.35) the value is written as the discounted benefits since I s ds is exactly the benefit over the time interval s s + ds . The process Z is illustrated in Figure 2. Survival model.34) are discounting e− t r . The integral is interpreted as follows: n t e− s t r I s ds = I t min Tx n t e− s t r ds (2. Now the elements of the prospective formulas can be written as follows: n−t Ex+t = Et e − n t n t n t r I n s t ax+t n−t = Et A1 n−t = Et x+t e− e− r I s ds dN s (2. dN s = 0 for t ≤ s ≤ n. N and I at any point in time is the remaining lifetime at time 0 corresponding to age x. This is the conditional expectation of the discounted benefit I n . dN s is the benefit at time s from one unit of a term insurance.34) s t r where Et denotes an expectation conditional on Z t = 0. such that dN s = 0 if s = Tx and dN s = 1 if s = Tx . which we denote by Tx . The underlying stochastic variable. n−t Ex+t . ax+t n−t . determining all the stochastic processes Z. n t e− s t r dN s = I t 1 − I n e− Tx t r The fundamental elements in Equations (2.34). Thus. i.1. and I is thus an indicator process for the insured to be alive.22 Technical reserves and market values 0 alive → 1 dead Figure 2.

We have specified an asset in which the insurer can invest. What we are really doing here is mathematical finance dealing with arbitrage-free prices and investment strategies in a particular financial market. we hesitate and discuss the real content of S 0 . if r is deterministic. and we find that the value at time t of one unit deposited at time s equals S 0 t /S 0 s . P t n .36) r Instead of rushing to insert a corresponding discount factor appropriately. Hereby. We assume that there exists a short rate of interest r such that d 0 S t = r t S0 t dt i. we buy P t n /S 0 t units of the asset S 0 . i. If S 0 is deterministic. it is a good idea to underpin the result by a n so-called arbitrage argument as follows.2. i. discounting and the conditional expectation are based on some principal considerations about what a value is and how it is determined. we have specified a financial market containing only one asset S 0 in which the insurer invests all payments.e. At time n the value of our .1 Absence of arbitrage In this section we introduce the principle of no arbitrage and discuss how discounting connects to this principle. and we denote by P t n its value at time t.4 Liabilities and principles for valuation 23 valuate. By introducing S 0 we have started the specification of this market. are easily found. A zero coupon bond is an asset which pays one unit at the terminal time n. we have that P t n = e− n t r Even if this may seem obvious. S0 t = e t 0 S0 0 = 1 (2. and bond prices and prices of options on bonds etc. On the other hand. We assume that payments to the insurer are deposited in a bank account with continuously accrued interest. Assume that P t n = exp − t r + . We denote by S 0 t the value at time t of one unit deposited in the account at time 0.e. in S 0 . 2. Negative payments in the form of benefits are withdrawn from the bank account. the other elements. we have not specified any investment alternatives.4. Now assume that we sell the bond and invest its price.e. If S 0 is deterministic. we can easily valuate deterministic payments.

in particular? Such considerations typically take the principle of no arbitrage as the theoretical starting point. or can issue. Certain areas of mathematical finance deal with questions such as: Given a stochastic model for r. what can be said about P t n in general and about its relation to P t u t < u < n. We do not accept the possibility of creating value without risk and conclude that = 0.24 Technical reserves and market values investments. is known at time 0. Thus. 2.4. The question is. Throughout this chapter r is assumed to be deterministic.4. Chapter 3 deals with stochastic interest rates. exp − 0 r I n cannot be interpreted as a value but as a stochastic present value.1 we stated that valuation of deterministic payments is simple if S 0 t is assumed to be deterministic. the time of death and other conditions which may determine a n payment in general are not known at time 0. what can be said about the value of a stochastic present value? A special situation arises if the insurer issues. This is the principle of no arbitrage. which is the benefit at time n. Denote by I i t the function indicating that insured number i is alive at time t. If S 0 is not deterministic. we simply obtain the value e− n 0 r I n However. is given by P t n S 0 n /S 0 t − 1 = e n t r If = 0 this value is created without any risk. the payment I n . for example.2 Diversification In this section we introduce the principle of diversification and discuss how conditional expectation connects to this principle. In Section 2. Then the law of large numbers comes into force and . Thus. an arbitrage argument deals with prices and investment strategies avoiding the possibility of risk-free capital gains beyond the interest rate. contracts to a “large” number m of insured with independent and identically distributed payments. it is by no means clear what can be said about P t n . after fulfillment of the obligation to pay one unit to the owner of the bond. If.

If death is assumed to occur with intensity . i. and particular asymptotic arbitrage arguments lead exactly to the value given in Equation (2. It is important to note that none of the presented principles. We mention that the mathematical financial term for the combination of the principles is the principle of no asymptotic arbitrage. we can put up the liability at time t. we obtain the following candidate to the value of I n at time 0: e− n 0 r E I n = e− n 0 r+ = n Ex (2. But even if one does not want to sell. one needs to set a price. By repeating the reasoning leading to the value n Ex .37) giving the liability at time 0 of a pure endowment insurance issued to an insured who at time 0 is x years old. .37) builds on the principles of no arbitrage and diversification. All these institutions are interested in the value of future obligations.37). one is interested in evaluating future payments at any point in time during the term of the contract. We say that the value in Equation (2. various institutions may be interested in the value of future payments. The value in Equation (2. Assuming that the insurer knows that the insured is alive at time t. For various reasons. solve the valuation problem by itself. we obtain the following value: e− n t r Et I n = e − n t r+ = n−t Ex+t giving the liability at time t of an insurance contract issued to an insured who was x years old at time 0 and who is alive at time t. 1 m m i=1 e− n 0 r I i n → e− n 0 r E I n as m → This result can be generalized to a situation when the insurance contracts do not have the same terminal time n.37) is a good candidate for the value at time 0 of I n paid at time n.e.2. regulatory authorities are interested in securing that the insurer can meet these payments with a large probability and put up solvency rules which are to be met. the tax authorities are interested in the surplus of the insurer as a basis for taxation.4 Liabilities and principles for valuation 25 we conclude that the total present value per insured converges towards the expected present value of a single contract as the number of contracts is increased. If one wishes to sell the obligation of the payments. The owners of the insurance company and other investors are interested in the value of the payments for the purpose of assessing the value of the insurance company itself. no arbitrage and diversification.

From Equations (2. and ∗ . We recall the interpretation of the elements.34).40) (2.24). (2. the following: V g t = Et n t n t r e− s t r bad dN s − I s I n n t r ds (2. Now consider the prospective versions of the technical reserve. We see that these quantities do not build on the principles of no arbitrage and diversification.28) and (2. the second order mortality rate. (2. In V ∗ t − V g t the payment process can be interpreted as continuous payments of a life annuity with time dependent annuity rates where the safety margin c t s makes up the annuity rate at time s ≥ t.26 Technical reserves and market values 2.41) e− e− s t r I s c t s ds s − c s I s ds s t r All these quantities are seen to build on the principles of no arbitrage and diversification.39) (2. respectively. as follows: V ∗ t = Et V ∗ t = Et∗ n t n t e− e− s t r bad dN s − I s bad dN s − I s ds + e− ds + e− n t r ba n I n ba t I n s ∗ t r n ∗ t r where the superscipts and ∗ on E denote that N under these expectations has the intensities . By a simple rewriting of the prospective market-based liabilities. Compared with the elementary payments in Equation (2.38) + b a t e− V b t = Et V ∗ t − V g t = Et V t − V ∗ t = Et b a n − b a t e− n t n t I n (2. but since the discount factor is not based on the bank account interest rate and since the intensity of N in the . The principles leave tracks in terms of discount factors and conditional expectations. the first order mortality rate.30) we obtain with help from Equation (2.3 The market-based liability revisited In this section we show how the principles of no arbitrage and diversification underpin the market-based composition of the liability. we can trace both discount factors and conditional expectations.4. the pension sum ba t and the premium appear in V g t . respectively.34) we see that the death sum bad . Indeed.22). The bonus payment ba n − ba t appears in V b t . The rate s − c s plays a corresponding role in the quantity V t − V ∗ t . we unveil the principles on which these are based.

42). given by Equation (2. these quantities can only be said to build on suitable imitations of the principles. The payments are the elements of the total payments in an insurance contract.33). The process Z serves to describe precisely a number of elementary claims and payment processes in life and pension insurance. for example monthly premiums. Fixing the time horizon n for the insurance contract. 2. since it describes payments floating between two parties. but there are many . The time of death is modeled by a probability distribution of the counting process N .43) shows the elements of these changes. It turns out to be beneficial to introduce the stochastic process Z based on the stochastic time of death.42) Here.39). and Equation (2.5 The liability and the payments In this section we introduce the idea of a payment process and discuss how the market-based composition is built from payment processes of guaranteed payments and payments which are not guaranteed. To formalize claims we introduce a payment process B such that B t represents the accumulated payments from the insurer to the insured over 0 t . In our example. This means that payments from the insured to the insurer appear in B as negative payments. t n is an indicator process for t ≥ n.43) t dB s 0 0≤t≤n (2. bad dN s ba t I n − I s ds and ba n − ba t I n The first task in connection with a precise description of the payments is to identify the stochastic phenomena on which the claims depend. We also speak of B as a payment stream. We specify the payments in continuous time. the payments are the elements of Equations (2. we can describe the claims in our example by collecting them in the payment process in the following form: B t = where dB t = bad dN t + ba t I t d t n − I t dt (2. for example by the introduction of a deterministic intensity . the integration sums up the infinitesimal changes dB.5 The liability and the payments 27 conditional expectation is not . For certain elementary insurances the time of death plays an important role. In Equation (2.38) and (2. although in practice the payments are discrete.2.

we search for the present value and the market value of a payment process. Premium waiver and other types of disability assurances can be described by extending the state space of Z with a third state.5. All generalizations of liability formulas are left to the reader. Disability model with recovery. one can introduce a J dimensional counting process where the jth entry N j counts the number of jumps into state j. let Z be a process moving around in a finite number of states J . The disability model is a three-state model. Models with more states are relevant for other types of insurances. we obtain the market value at time t of a payment process as follows: n Et e− s t r dB s (2.28 Technical reserves and market values 0 active → (←) 2 dead 1 disabled Figure 2. and we get the present value at time t of the payment process B over t n as follows: n t e− s t r dB s Combining the principles of no arbitrage and diversification. J = 3.2.1 The market-based liability revisited In this section we identify the payment processes in the market-based composition of the liabilities. situations which are not described by such a process. “disabled.2. With the collection of claims in a payment process. We mention this to give an idea of how the construction of payment processes generalizes to various forms of disability payment processes etc. One example is premium waiver. in general. i. The situation with a disability state is illustrated in Figure 2. for example contracts on two lives where each member of a couple is covered against the death of the other. The principle of no arbitrage determines the present value as a sum of present values of the single elements of the payment process.e. . Corresponding to a general J -state process Z. An important reference is Norberg (2000).44) t 2.” One can.

5 The liability and the payments 29 In Equations (2.45) We remind ourselves that this constitutes the individual bonus potential if U t ≥ V∗ t ≥ V t . We now collect these claims in payment processes such that all entries can be written as market values of payment processes. which we denote by B t · . We introduce now the guaranteed payment process at time t.32). the suggested definition of the individual bonus potential in Equation (2. for s ≥ t. such that for s ≥ t we have dC t s = I s c t s ds V ∗ t − V g t = Et n t e− s t r dC t s (2. It is suitable to characterize the entries in the balance sheet by different payment processes. We also have in mind the situation where the condition X n = 0 cannot be fulfilled. denoted by C t · . This makes it easy to generalize the entries to other insurances by simply specifying the generalized payment processes. we denote by Bb t · the payment process not guaranteed at time t.2. and. since Equation .38)–(2. dB t s = − I s ds + bad dN s + ba t I s d V g t = Et n t s n e− s t r dB t s Correspondingly. We end this section by repeating. for the case U t ≥ V ∗ t . Then for s ≥ t. dBb t s = ba s − ba t I s d V b t = Et n t s n e− s t r dBb t s The difference V ∗ t − V g t stems from the accumulated safety margins in the guaranteed payments.41) we wrote the entries in the market-based composition of the liability in terms of claims.

6. the quantities in Equations (2.46) which is not based on the future safety margins.48) comprise alternatives to the definition given in Equation (2. Thus. In Section 2. Survival model with surrender option.47) e− r dC + t s (2.3 we suggest a third alternative to Equation (2. The definition of the individual bonus potential could be based on a more precise description of how the safety margins are used for bonus.48) are closely connected to the discussion about what an interest rate guarantee is worth. The two alternatives coincide with the first definition if c t s (see Equation (2. 2.29)) is either positive or negative for all s.48) is connected to a certain interpretation of the contribution principle.6.48) Here.48) are the basis for calculation of prices on interest rate guarantee options. Values similar to Equation (2. Equation (2.30 Technical reserves and market values (2. The quantities in Equations (2. Of course.45) inspires us naturally to suggest two alternative definitions. Whereas Equation (2.47) is connected to a terminal bonus. we approach the value of the surrender option by the threestate Markov model illustrated in Figure 2.3.3.6 The surrender option 2. we introduce a state of 0 alive and insured → 1 surrender 2 dead Figure 2. we have the following inequalities: V ib t = Et ≤ Et ≤ Et n t n t n t e− e− s t s t + r dC t s + (2.1 Intensity-based surrender option valuation In this section.46) s t r dC t s (2.46).47) and (2.47) and (2. .

52) + e− s t r+ + r+ t −c ≤ V∗ t −V t p t where p t = n t e− s t s ds Secondly. One now obtains the following: V sur t − V t = n t e− n t s t r+ + s e s s t r+ s V∗ t −V t e s ds d ds (2. we consider the case where V t ≥ V ∗ t . one is interested in a simplified estimate of the value of the surrender option. It can be verified that this additional reserve can be written in the following form: V sur t − V t = n t (2.e.e. i.51) Firstly. i. We consider the case where the surrender value equals the technical reserve. which at time t is reached by a surrender intensity and leads to a payment of a surrender value G t .6 The surrender option 31 surrender. We assume that the future second order basis is settled such that X s ≥ 0 for . i. We assume that the future second order basis is settled such that s ≤ c s for s > t. V sur − V . X t < 0.e.2. Therefore. Using a version of Thiele’s differential equation for a three-state model. G t = V ∗ t . such that V sur t − V t = n t e− s t r+ + s V∗ s −V s ds (2. we consider the case where V t < V ∗ t .49) e− s t r+ + s G s −V s ds (2. i.e. we can write the differential equation with terminal condition for the total reserve V sur including the surrender option as follows: d sur V t = r t V sur t + − t bad − V sur t dt − t G t − V sur t V sur n = ba n We require an explicit expression for the value of the surrender option. X t ≥ 0.50) The calculation of V sur t − V t involves the future second order basis through future values of G and V .

One problematic circumstance is that there seems to be only weak historical evidence from situations where surrender was optimal. One can certainly criticize the approach taken in this section. By introducing the surrender option. depending on what occurs first.6.53). given the constraints on the future second order basis leading to Equations (2. given a situation where surrender is optimal. we have the following inequality: V sur t − V t ≤ max p t G t − V t 0 2.32 Technical reserves and market values s > t. Here. we can think of B as the payment process for the example given by Equation (2.3. Will we in practice allow for a situation where the insured gains on surrender? In this section the starting point is that the insured surrenders immediately if it is advantageous to do so. One speaks . Consider a general payment process B. which should be consulted for a generalization of the payment processes. Another circumstance is that.” Hereby we take into consideration the fundamental difference between transitions between the states in the model illustrated in Figure 2.2 Intervention-based surrender option valuation In this section. the ideas and the results in this section.53) ≤0 We conclude that.54) where we put G n = 0.52) and (2. This can happen by either not allowing surrender or by making surrender less attractive by introducing costs. We then obtain the following inequality: V sur t − V t = n t e− s t r+ + s V∗ s −V s ds (2.43). an arbitrage argument gives the market value including the surrender option as follows: V sur t = max Et t≤ ≤n e− t s t r dB s + e− t r I G (2. Optimal strategies of the insured provide a starting point for a detailed study of the problem and a derivation of the corresponding deterministic differential systems given by Steffensen (2002). we approach the value of the surrender option by considering the idea that the insured surrenders “if it is worth it. and is the time of surrender or the terminal time. the insurer or the regulatory authorities will probably put up a protection against systematic surrender. Obviously the insured must decide on surrender at time t based on the information available.

Y is a martingale if. for u ≥ t. a martingale is defined as a process which is both a sub-martingale and a super-martingale. Et y Y u ≤ y where the subscript t y denotes that an expectation is conditional on Y t = y . If G ≤ V ∗ . we use the notions of a super-martingale and a submartingale. In the following.2. Y is a sub-martingale if. Formally. Y sur t u is the present value of the payments to the insured including the surrender value. Correspondingly. i. corresponding to the situation without surrender option. for u ≥ t. we see that the classical solvency rule U ≥ V ∗ exactly prevents the policy holder from making gains on surrender.1 (1) If V t ≥G t for all t. we consider a process Y sur t u defined by Y sur t u = u t e− s t r dB s + e− u t r I u G u Thus. it cannot be beneficial to surrender and the company should set aside V . a process which is expected to maintain its value over time. Et y Y u = y Proposition 2. and we explain briefly these notions in the Markovian case. A super-martingale describes the process that over time is expected to decrease in value compared to where it is.e. it is optimal never to surrender and V sur t = V t Intuition: if the total liability exceeds the surrender value. given that the contract is surrendered at time u. Formally. We recall that I t indicates whether the insured is alive at time t and that G t is the surrender value at time t. the condition for Y to be a supermartingale is that. a sub-martingale describes a process expected to increase in value.6 The surrender option 33 of as a stopping time. In order to say something general about the quantity V sur t . for u ≥ t. Formally. Et y Y u ≥ y Finally. Thus. the company gains by surrender the value V − G. .

V t and Y sur t u replaced by V g sur t . the policy holder should at any point in time postpone the surrender in order to increase the expected present value of the contract.1 with V sur t . it is never optimal to surrender. In Equation (2. Thus.34 Technical reserves and market values (2) If Y sur t u is a sub-martingale. it is optimal to surrender immediately and V sur t = G t Intuition: that Y sur t u is a super-martingale means that the present value of payments to the policy holder is expected to decrease as a function of the time to surrender. Thus. G t is the surrender value at time given that no dividends are distributed over t . V g t . . (3) If Y sur t u is a super-martingale. V b sur . and a reserve for the payments which are not guaranteed. and V sur t = V t Intuition: that Y sur t u is a submartingale means that the present value of payments to the policy holder is expected to increase as a function of the time to surrender. for example given by Equation (2. Y g sur t u is the present value of the guaranteed payments to the insured including the surrender value given that the contract is surrendered at time u. we need to identify whether Y g sur t u is a super-martingale or a sub-martingale. Now we can state Proposition 2. V g sur . and Y g sur t u . Given a value of the total payment process. We introduce the process Y g sur t u as follows: Y g sur t u = u t e− s t r dB t s + e− u t r I u G t u Thus. the policy holder should surrender immediately in order to obtain the highest possible expected present value of the contract. the question is how to decompose this total reserve into a reserve for the guaranteed payments.55) and then determine V b sur residually as V sur −V g sur .55).54). In order to apply this result. One idea is to define V g sur as follows: V g sur t = max Et t≤ ≤n e− t s t r dB t s + e− t I G t (2.

we have that V g sur t = V g t Intuition: the policy holder has the negative safety margins covered by keeping the insurance contract since we disregard all bonus payments.3 we put up specific conditions under which it is relatively easy to calculate the relevant maximum.3 (1) If c t s ≥ 0. The policy holder gains maximally in expectation by keeping the policy until termination. then the process itself is a super-martingale or a sub-martingale. Corollary 2. for calculation of the general market values V sur and V g sur . Nevertheless.1 and Corollary 2.6 The surrender option 35 respectively. Then it is possible to write Y g sur t u as follows: Y g sur t u = V ∗ t − u t e− s t r I s c t s ds + M sur t u where M sur t s is a martingale. such that the value of the guaranteed payments simply becomes the corresponding value without the surrender option.2 Assume that G t u = V ∗ t u .2. This can be avoided by immediate surrender such that the value of the guaranteed payments simply becomes the surrender value. Even though this goes beyond the scope of this chapter. In Proposition 2. it is possible to derive a deterministic differential system. we have that V g sur t = V ∗ t Intuition: the policy holder loses the future positive safety margins by keeping the contract. comparable with Thiele’s differential equation. Lemma 2. This result is applicable once we have established the following lemma. (2) If c t s ≤ 0. we . We can now conclude the following for the case G t u = V ∗ t u as follows. We use a result stating that if a process can be written as the sum of a decreasing or increasing process and a martingale. respectively.

31) for the case V ≥ V ∗ . The equality in the third line is the formalization of the following statement: at any point in the state space. one of the two inequalities is an equality. We have that V g sur t = V g sur t t . The second inequality simply states that the sum at risk connected to the transition to the state surrender is always negative.56) coincides with the definition in Equation (2.6. where V g sur t u ≤ + r u V g sur t u − u bad − V g sur t u u V g sur t t ≥ G t 0= − − V g sur t n = ba t The reader should recognize several elements in this system.47) and (2. We repeat the definition of the individual bonus potential given in Equation (2. since the studies in Section 2.3 The market-based liability revisited Corollary 2.6.3 is closely connected to the market-based composition of the liabilities.48) we gave alternatives to the definition in Equation (2. out of interest.2 lead us.31) if c s t is either positive or negative for all s.31) in connection with a discussion of how the safety margins are redistributed as dividends.56) we suggest an alternative which does not contain this subjectivity of the dividend distribution. together with the terminal condition. u V g sur t u + + r u V g sur t u × G t u − V g sur t u u bad − V g sur t u 2. Corollary 2. . We have the following inequality: V ib = max V ∗ V g − V g ≤ V g sur − V g (2. In Equation (2. only differs from Thiele’s differential equation by containing an inequality instead of an equality.36 Technical reserves and market values give. to an alternative definition. the result for V g sur .3 shows us that the alternative definition in Equation (2. The first inequality. In Equations (2.56) with G t u = V ∗ t u . quite naturally.

We now calculate this bonus potential and start 0 alive and with premium → 1 free policy 2 dead Figure 2. Survival model with free policy option.1.7 The free policy option 37 Actually. given the free policy option. the maximum is taken over all points in time between t and n. . it is the inequality in Equation (2.1 A simple free policy option value The starting point for the value of the free policy option is a model with three states as illustrated in Figure 2. the idea is not to introduce a free policy intensity in the same way as we introduced a surrender intensity in Section 2. The liability concerning bonus stemming from the future premiums is given by the bonus potential on a contract with premium issued at time t with a deposit equal to zero.4. However. One can argue that if we take into account the future premiums we should also. We recall that V∗ t −Vg t = n t e− s t r+ c t s ds A part of the safety margin in the guaranteed payments can be said to stem from future premiums.56) follows from max V ∗ V g ≤ V g sur which is true since the left hand side corresponds to Equation (2.55).4. put up a reserve for the bonus based on these premiums. where the maximum is taken over only two possible points in time.7 The free policy option 2. The inequality in Equation (2. t and n. 2.2. On the right hand side.56) which motivates us to suggest the alternative definition. for the individual bonus potential.7. V g sur − V g .6.

We reach the following: V ∗ t V g+ t −Vg t V ∗+ t +Vg t = V ∗ t V g+ t ≡ Vf t V ∗+ t . by Equations (2.58) i.57) and (2.60) The number of units of guaranteed benefits.60) and (2. t .59) (2. the superscript + denotes here that only the benefits are included. that can be bought from the future premiums at time t are defined simply by the equivalence principle: t a bad A1∗ x+tn−t + b t ∗ n−t Ex+t = a∗ x+tn−t which.57) (2. (2.22).61) In Section 2.59).2 we noted that the bonus potential on an insurance contract at the time of issuance simply equals the negative market value of the guaranteed payments stipulated in the contract.61) equals V ∗ t V g+ t −Vg t V ∗+ t This quantity is now our candidate for the value of the free policy option.e. from Equations (2.9) and (2. Then we can. We see that the bonus potential on the new insurance is given by − t a bad A1 x+tn−t + b t n−t Ex+t + ax+tn−t which by Equations (2. write the life annuities a∗ x+tn−t and ax+tn−t in the following way: a∗ x+tn−t = ax+tn−t = V ∗+ t − V ∗ t V g+ t − V g t (2.38 Technical reserves and market values out by finding the benefits on the issued contract. We introduce the following notation: a V ∗+ t = bad A1∗ x+tn−t + b t a V g+ t = bad A1 x+tn−t + b t ∗ n−t Ex+t n−t Ex+t (2.3.58). yields t = V ∗+ t − V ∗ t V ∗+ t (2. This value must be added to the market value of the guaranteed original payments in order to obtain the value of the guaranteed payments including the free policy option.

2 closely.4.7.7 The free policy option 39 Note that V f is also the market value of the guaranteed free policy benefits since the guaranteed benefits upon conversion into a free policy are exactly reduced by the factor V ∗ t /V ∗+ t .2. It turns out to be informative to decompose the individual bonus potential into the bonus potential on premiums V bp and the bonus potential on the free policy V bf .” Hereby we take into consideration the fundamental difference between transitions between the states in the model illustrated in Figure 2. A situation can arise where these safety margins are negative.6.6.2 Intervention-based free policy option valuation In this section the free policy option is dealt with in the same way as the surrender option was dealt with in Section 2. can also be expressed as the market value of the safety margins in these payments.62) If. the bonus potential on premiums is in this case set to zero. Since negative safety margins cannot be distributed as bonus. we want to take into consideration all possible relations between V . V bp t . One problem is that we have no historical evidence from a situation where . For the case V ≥ V ∗ we obtain an expression for V bp which is similar to Equation (2. at one time. Even though the intuitive interpretations on this background could be left to the reader. We now take the logical starting point that the insured converts into free policy “if it is worth it.31): V bp = max V f V g − V g (2. we keep them here and follow Section 2. Firstly.2. we consider the situation V ≥ V∗ ≥ Vf ≥ Vg Then V ib = V ∗ − V g decomposes into V ib = V bf + V bp = V∗ −Vf + Vf −Vg The market value of guaranteed payments on a contract issued at time t. One can certainly criticize the approach taken in this section. V f and V g . the bonus potential on premiums and the bonus potential on the free policy can be formalized as follows: V bf = V − V g − V f − V g + + − V −Vg − V∗ −Vg + + + + V bp = V − V g − V − V g − V f − V g 2. V ∗ .

one cannot stop conversions into free policy.e. Another is that. however. In opposition to the surrender option. which should be consulted for a generalization of the payment processes. the insurer or the regulatory authorities will probably put up a protection against systematic conversion.40 Technical reserves and market values conversion into a free policy was optimal.63) where V free corresponds to V sur with the surrender value G t replaced by the reserve V f + Xf . which we define by Y free t u = where Vf t = V ∗ t V g+ t V ∗+ t u t e− s t r dB s + e− u t r I u V f u + Xf u Thus. i. will one in practice allow for a situation where the insured gains on conversion? In this section the starting point is that the insured converts immediately if it is advantageous to do so. and is the time of conversion or the terminal time depending on what occurs first. an arbitrage argument gives the market value including the free policy option as follows: V free t = max Et t≤ ≤n e− t s t r dB s + e− t r I Vf + Xf (2. X f = X. if the full undistributed reserve is carried over upon conversion. is a stopping time. given a situation where conversion into free policy is optimal. By introducing the free policy option. We think of B as the payment process for the example given by Equation (2. Optimal strategies of the insured provide a starting point for a detailed study of the problem and a derivation of the corresponding deterministic differential systems given by Steffensen (2002). This would involve difficult considerations about credit risk. The quantity X f is the undistributed reserve upon conversion. given that conversion happens at time u. Thus. Thereby. the ideas and the results in this section. including the liability to cover free policy benefits and benefits which are not guaranteed from the time of conversion. This can happen by making conversion less attractive by introducing costs. . Obviously the insured must decide on conversion at time t based on the information existing at that time. Consider a general payment process B. we consider a process Y free t u as a function of u. Y free t u is the present value of the payments to the insured.43). In order to say something general about the quantity V free t .

2. Thus.7 The free policy option 41 Proposition 2.4 (1) If V t ≥ V f t + Xf t for all t.63). for example given by Equation (2. Thus. we see that the classical solvency rule U ≥ V ∗ exactly prevents the policy holder from making gains on conversion. corresponding to the situation without free policy option. (2) If Y free t u is a sub-martingale. the company gains by conversion the value V − V f t + X f t . it cannot be beneficial to convert and the company should set aside V . (3) If Y free t u is a super-martingale. V b free . If V f + X f ≤ V ∗ . the policy holder should at any point in time postpone the conversion in order to increase the expected present value of the contract. the policy holder should convert immediately in order to obtain the highest possible expected present value of the contract. Given a value of the total payment process. We suggest the definition V g free t = max Et t≤ ≤n e− t s t r dB t s + e− t I Vf t (2. it is optimal to convert immediately and V free t = V f t + X f t Intuition: that Y free t u is a super-martingale means that the present value of payments to the policy holder is expected to decrease as a function of the time to conversion.64) . it is never optimal to convert and V free t = V t Intuition: that Y free t u is a sub-martingale means that the present value of payments to the policy holder is expected to increase as a function of the time to conversion. V g free . the question is how to decompose this total reserve into a reserve for the guaranteed payments. Thus. and a reserve for the payments which are not guaranteed. it is optimal never to convert and V free t = V t Intuition: if the total liability exceeds the market value of payments at conversion.

including the liability for future free policy benefits. See Steffensen (2002) for details of the function cfree t u . This can be avoided by immediate conversion such that the value of the guaranteed payments simply becomes the market value of the free policy benefits. Now we can state Proposition 2. we have that V g free t = V f t Intuition: the policy holder loses the future positive safety margins on the future premiums by continuing to pay the premiums on the contract since we disregard all future bonus payments.6 (1) If cfree t u ≥ 0. V f t +X f t and Y free t u replaced by V g free t . Corollary 2.42 Technical reserves and market values where Vf t = V g+ t V∗ t V ∗+ t Then one could choose to define V b free residually by V b free = V free − V g free . . This result is applicable once we have established the following lemma. We introduce the process Y g free t u given by Y g free t u = u t e− s t r dB t s + e− u t r I u Vf t u Thus. Y g free t u is the present value of the guaranteed payments to the insured.5 It is possible to write Y free t u as follows: Y free t u = V f t − u t e− s t r I s cfree t s ds + M free t u where M free t u is a martingale and cfree t u is a rather involved function of V ∗ t . respectively. given that the contract is converted at time u. In order to apply this result. respectively. V g t and Y g free t u . We use a result stating that if a process can be written as the sum of a decreasing or increasing process and a martingale. We can now conclude the following. Lemma 2. V g t .4 with V free t . we need to identify when Y g free t u is a super-martingale or a sub-martingale. V ∗+ t . V g+ t and the first order and real bases. then the process itself is a super-martingale or a sub-martingale.

we give.7 The free policy option (2) If cfree t u ≤ 0. the result for V g free . one of the two inequalities is an equality.7.6 is closely connected to the market-based composition of the liabilities. In Proposition 2. We repeat the definition of the individual bonus potential on premiums given in Equation (2. since the studies in . out of interest.6 we put up specific conditions under which it is relatively easy to calculate the relevant maximum. such that the value of the guaranteed payments simply becomes the corresponding value without the free policy option.2. The policy holder gains maximally in expectation by continuing to pay the premium on the policy until termination. The first inequality together with the terminal condition.31) for the case V ≥ V ∗ . only differs from Thiele’s differential equation by containing an inequality instead of an equality. comparable with Thiele’s differential equation. We have that V g free t = V g free t t . u V g free t u + + r u V g free t u × V f t u − V g free t u u bad − V g free t u 2. The second inequality simply states that the sum at risk connected to the transition to the free policy state is always negative.4 and Corollary 2. The equality in the third line is the formalization of the following statement: at any point in the state space. Nevertheless.3 The market-based liability revisited Corollary 2. it is possible to derive a deterministic differential system. where u V g free t u ≤ + r u V g free t u − u bad − V g free t u V g free t t ≥V f t 0= − − V g free t n = ba t The reader should recognize several elements in this system. for calculation of the general market values V free and V g free . Even though this goes beyond the scope of this chapter. we have that V g free t = V g t 43 Intuition: the policy holder has the negative safety margins on future premiums covered by continuing to pay the premiums on the insurance contract since we disregard all bonus payments.

We have the following inequality: V bp = max V f V g − V g ≤ V g free − V g (2. where the maximum is taken over only two possible points in time. . t and n. the maximum is taken over all points in time between t and n. The inequality in Equation (2.44 Technical reserves and market values Section 2. it is the inequality in Equation (2.2 lead us.6 shows us that the alternative definition in Equation (2.7.62) if cfree t s is either positive or negative for all s. quite naturally.64).65) coincides with the definition in Equation (2. V g free − V g . to suggest an alternative definition. for the bonus potential on premiums. On the right hand side.65) Corollary 2.65) follows from max V f V g ≤ V g free which is true since the left hand side corresponds to Equation (2.65) which motivates us to suggest the alternative definition. Actually.

The present chapter is organized as follows.5. this section demonstrates how versions of Thiele’s differential equation can be derived for the market value of the guaranteed payments.3 gives a more systematic treatment of topics such as zero coupon bonds.6 we briefly turn to the estimation of forward rates. the analysis provides a basis for an integrated description of the liabilities and 45 . There exists a huge amount of literature on financial mathematics and interest rate theory. we mention more general bonds and relations to zero coupon bonds. Section 3. This argument. to the case of stochastic interest rates.2 demonstrates how the traditional actuarial principle of equivalence can be modified in order to deal with situations with random changes in the future interest rate. This discussion leads to formulas for market values involving so-called martingale measures. leads to new insights into the problem of determining the market value for the guaranteed payments on a life insurance contract. and we shall not mention all work of importance within this area. which involves hedging via so-called zero coupon bonds.e. In these equations. Section 3. and in Section 3. In Section 3. the term structure of interest rates and forward rates. Readers interested in more mathematical aspects of these theories are referred to Lamberton and Lapeyre (1996) and Nielsen (1999). In addition. Some basic introductions are Baxter and Rennie (1996) and Hull (2005). Finally we mention Björk (1997. where trading is possible at fixed discrete time points. i. Section 3.1 Introduction This chapter provides a brief introduction to some basic concepts from interest rate theory and financial mathematics and applies these theories for the calculation of market values of life insurance liabilities. 2004) and Cairns (2004).7 presents an introduction to arbitrage pricing theory in models.3 Interest rate theory in insurance 3. forward rates now appear instead of the interest rates. In addition.

bond prices and stock prices.46 Interest rate theory in insurance assets of the insurance company which can be used for assessment of the company’s total financial situation under various economic scenarios. 3.2. since in this case the market value of the liabilities is also affected by the financial situation. 3. the average number of heads from tossing the same coin m times) will converge to some number as the number of experiments increases.2 Valuation by diversification revisited How are life insurance contracts traditionally valuated? What is the reasoning behind these methods? What is diversification? To answer these questions. Essential elements are stochastic models for the future interest rate. .2.” i. The law of large numbers states that the average from some experiment (for example.e. we could let X i be the outcome from the ith toss.2 A portfolio of insured Consider now a portfolio of lx . For t ≥ 0 we denote by lx+t the expected number of survivors at age x + t. We impose here the following standard assumptions. where Ai is the event that the ith toss leads to a “head.1 The law of large numbers A first version of the law of large numbers was formulated by the Swiss mathematician Jakob Bernoulli. Since the expected value of the indicator function 1A1 is equal to the probability of the event A1 .” We can write this in a more compact form using the indicator function X i = 1Ai . which for most coins 1 should be 2 . Ai = ith coin head . we consider situations with stochastic interest. say. In the coin-tossing example. The importance of this analysis is increased by the introduction of market-based accounting in life insurance. identical n-year pure endowment contracts with sum insured 1. we see that the average of heads converges to the probability of observing a head. 3. In particular. by taking X i = 1 if the coin shows a “head” and X i = 0 if it shows a “tail. let us first recall the traditional actuarial law of large number considerations and see how these can be modified for the most simple life insurance contracts in the presence of financial uncertainty. Then the law of large numbers states that 1 m m 1Ai → E 1A1 i=1 (3.1) as m increases to infinity.

We emphasize that the assumed independence is crucial in order to obtain this convergence. In more realistic models. The ratio between the actual (unknown) number of survivors at time n and the number of policy holders lx entering the contract at time 0 can now be written as follows: 1 lx lx 1 T i >n i=1 (3. we know that if the portfolio is large.3) The situation is now almost identical to the coin-tossing example. This shows that the actual number of survivors is indeed close to the number predicted by the decrement series lx+n if lx is “big.2 Valuation by diversification revisited 47 • The lx policy holders are all of age x at time 0 with remaining lifetimes given by T 1 T lx with the same survival probability. we cannot predict exactly the number of policy holders. the law of large numbers does not imply that . and where we have used standard actuarial notation. at which time each survivor receives one unit. provided that the lifetimes of the policy holders are independent.3) converges to E 1 T 1 >n = P T 1 > n = n px = lx+n lx as the number of policy holders lx increases to infinity. could be a Gompertz–Makeham intensity of the form x+t = + cx+t 0 at time 0.” i. However. For example. Equation (3.e. introduce the indicator 1 T i >n that the ith policy holder survives. To see this. by the law of large numbers. who actually survive until time n. t px = lx+t = exp − lx t 0 x + u du (3.3. • The contracts are paid by a single premium Of course. where the future mortality intensity is unknown (stochastic). Similar arguments can be applied in situations where several time points or even payment processes are considered. this is a consequence of the law of large numbers. lx i=1 1 T i >n ≈ lx n px = lx+n In a sufficiently large insurance portfolio. the actual number of survivors at a given fixed time is hence relatively close to the expected number of survivors. lx .2) where the mortality intensity is a deterministic function. the actual number of survivors at time n is in some sense close to the expected number lx+n . Mathematically. and.

and the unsystematic risk is the risk associated with the insured lifetimes given the underlying mortality intensity.3 Interest. The quantity r is known as the force of interest. How can this be generalized to situations where i is no longer constant? One possibility is to let i s represent the annual interest rate for year s. Here. This problem is addressed further in Section 5. The t-year discount factor is given by vt = 1 + i −t = e−rt = S t −1 (3.3) converges to a constant. and that this rate is constant during the term of the insurance contract. and it can be interpreted as the interest per time unit per unit deposited on the account. it is typically assumed (implicitly) that the insurance company invests capital in an account with yearly interest rate i. In the traditional actuarial literature. the t-year accumulation factor is determined as follows: 1 + i t = ert = S t which clearly satisfies the differential equation d S t = rS t dt (3.5) (3. accumulation and discount factors What is the relation between yearly interest rate and the force of interest? How do we handle interest rates that are not constant during the term of the contracts? In this section. Thus. we can alternatively write the one-year accumulation factor 1 + i as er . one distinguishes between unsystematic and systematic mortality risk. and the t-year accumulation factor is given by t S t = 1+i 1 ··· 1+i t = exp s=1 rs (3.6) where v = e−r .5. The systematic risk is associated with the consequences of random changes in the underlying mortality intensity.4) with the initial condition S 0 = 1. whereas the systematic is undiversifiable and remains with the insurer.2. see for example Gerber (1997).7) . The corresponding force of interest r s for year s is then determined from the equation 1 + i s = er s . which is diversifiable and can be eliminated by increasing the size of the portfolio. It is the unsystematic risk.48 Interest rate theory in insurance the quantity (3. Introducing r = log 1+i . we recall these and other basic concepts. 3.

9) and the corresponding discount factor is given by St −1 = exp − t r u du 0 (3. This extension is the basis for more realistic studies of the impact of changes in the interest rates on the balance sheet of an insurance company. an area which is essential within asset-liability management. that is d S t =r t S t (3. one refers to this account as a savings account. this framework can be used when building models which allow for random changes in the future interest rates. then the force of interest is the most natural starting point. We use the accumulation and discount factors from Equations (3.10) in the following. we can define a new accumulation factor by changing Equation (3.4 The insurer’s loss and the principle of equivalence Now consider a portfolio of lx pure endowments of one unit expiring at time n paid by a single premium 0 at time 0. 3.2 Valuation by diversification revisited 49 If we consider instead the case where the interest rate can change more often than once every year. and the present value of the premiums is simply equal to the payment lx 0 .10) Note that the quantity S t represents the value at time t of one unit invested at time 0 in the account which bears continuously added interest r u .5). each of the survivors at time n receive one unit.9) and (3. Letting r u be the force of interest at any time u. but if we solve Equation (3. In financial mathematics. no discounting is needed. the present value at time 0 of this benefit is given by S n −1 lx+n .8) with the initial condition S 0 = 1. the present value of the insurer’s loss associated with the portfolio can be defined as follows: L = lx+n S n −1 − lx 0 (3. Thus.11) . In addition. One advantage compared to working with a constant interest rate is that we can now allow interest rates to vary over time by specifying the force of interest as a function of time. With this contract. Assume moreover that the number of survivors follows the decrement series.3. i. since it describes the interest per time unit.e. it is deterministic and equal to lx+n .5) such that r depends on time. the accumulation factor becomes S t = exp t r u du 0 (3.2.8) dt It does not look like a big difference compared to Equation (3. Since the premiums lx 0 are payable at time 0.

the price at time 0 ≤ t ≤ n is denoted by P t n . However.e. if L = 0.) A zero coupon is a contract which pays its holder one unit at a fixed time n (also referred to as an n-bond). we see from (3.12) that we did not get rid of the factor S n −1 in the case where S n is random. we cannot charge this premium. In this situation. where tomorrow’s prices for zero coupon bonds are known today. This is not so surprising. Before analyzing the company’s loss in this setting. However.2. i. it is important to realize that this argument works only in the situation where n the discount factor S n −1 = exp − 0 r u du is deterministic.3 and 3. 3. If S n is deterministic (or known at time 0). In the case where the future interest rates are unknown at time 0. we also say that mortality risk is diversifiable. i. but it is nevertheless worth pointing out. In this setting.e. we get from Equation (3. Clearly.11) the following well known result: 0 = lx+n exp − lx n 0 r u du = n px exp − n r u du 0 (3. the situation with deterministic bond prices. we must have that P n n = 1. . since it raises the question of how we can control or eliminate this risk.50 Interest rate theory in insurance The premium 0 is now said to be fair. the fair premium is the well known equivalence premium: t using the fact that S t = exp 0 r u du . whereas the financial risk related to the future development of the interest rate could not be eliminated by increasing the size of the portfolio. we see from the simple considerations above that mortality risk can essentially be eliminated by increasing the size of the portfolio if the insured lives are independent. i. we consider the special case where all future zero coupon bond prices are known at time 0. The answer is straightforward in the idealized world of our simple example: by trading with financial contracts called zero coupon bonds.e. (Sections 3. we can derive the structure of zero coupon bond prices and give another introduction to the concept of arbitrage.5 below are devoted to a more systematic treatment of this topic. there is no randomness. since we do not know S n at the time of selling of the contract! Thus. if there is no systematic insurance risk within the model.12) which is simply the expected present value of the benefit.5 Deterministic bond prices Let us consider a completely deterministic world. We can interpret this by saying that we have eliminated the mortality risk.

14) and that we can invest in the savings account with deterministic interest r u as explained in Section 3. Thus.14) An alternative way of addressing this issue is as follows. For reasons of completeness. This shows that P t and P n can be interpreted as discount factors. .2. At time . it is essential to realize that Equation (3. However.3. which can be used to buy an n-bond. this leads to the payment P n . this implies that P t n =P t P n (3. More precisely. It can now be shown that if we insist on prices which do not allow for the possibility of risk-free gains. recall that P t n is the price at time t of one unit at time n.13) implies the existence of a function r ∗ such that P t n = exp − n t r ∗ u du (3. one can actually prove that Equation (3.13) which says that the value at time t of an n-bond is equal to the value at time t of a -bond multiplied by the value at time of an n-bond. when > t. We have thus constructed a market with two investment possibilities (two assets).2 Valuation by diversification revisited 51 Fix some times t ≤ n.15) so that r ∗ u . must be identical to the interest rate r u from the savings account. the price P n at time of an n-bond is known already at time t for any ∈ t n . In particular. Assume that we can buy zero coupon bonds with maturity n at the price given by Equation (3. Under the assumption of deterministic zero coupon bonds. respectively. then there is only one possible price for the zero coupon bond. To show that Equation (3. The reason is again that P t n and P t are known at time t. one can alternatively obtain one unit at time n by investing in P n -bonds at the price P t P n . and this ensures the payment of one unit at time n.13) is satisfied. Assuming in addition that the zero coupon prices are sufficiently nice (smooth) functions. whereas P n is not known at t. we show that the only reasonable price for the bond is given by P t n = exp − n t r u du = St Sn (3.13) is not satisfied in the more realistic situation where P n is not known at time t for > t. which appears in Equation (3. If bond prices are deterministic. where t is today and n is the payment time. one can in this case think of P t and P n as discount factors for t and n .14). we give this argument here.3. This type of argument is also called an arbitrage argument.

We consider in this section only the guaranteed payments as defined in Chapter 2. This leads to the liability 1 at time n. we modify the principle of equivalence to the situation where the insurer can trade zero coupon bonds. i. This strategy leads to a gain of . and there would be no buyers. the deposit on the savings account is then P t n S n /S t = 1 + . the risk associated with the development of the interest rate. . which involves the future (unknown) interest rate. At time n. Consequently. This gives a way of hedging.15). Finally.6 Hedging with zero coupon bonds In this section. 3. The main idea of arbitrage-free pricing is that such possibilities cannot exist in the market. since all investors would want to sell the bond if > 0. Equation (3. This can be seen directly by considering Equation (3. It is now possible to construct a risk-free gain by using the following strategy (if > 0). as we see below. the treatment of payments related to bonus is postponed until Chapter 4. • At time n.15) simplifies to P t n = exp −r n − t which is identical to the classical discount factor given in Equation (3. there is no reasonable argument which supports this idea. One alternative idea could be to replace S t /S n by its expected value. withdraw the amount 1 + from the savings account and pay one unit to the buyer of the bond. If < 0. controlling or eliminating. However.2.e. • Sell one bond at the price P t n at time t.15) and is given by P t n = 1+ exp − n t r u du = 1 + St Sn for some = 0. • Invest the amount P t n in the savings account at time t. prices would adapt such that it is no longer possible to generate risk-free gains. a gain of − can be obtained by borrowing money from the bank and buying a bond. we note that in the case where r u is constant and equal to r. We point out that this argument cannot be applied in the situation where r u is not deterministic.6).52 Interest rate theory in insurance Assume that the price of an n-bond at time t differs from that given in Equation (3.

In the portfolio with lx policy holders. Hence.2 Valuation by diversification revisited 53 Hedging a portfolio of pure endowments Assume that the insurer at time 0 invests in lx units of an n-bond. In addition.17) has a very natural form: it is the price at time 0 of a zero coupon bond with maturity n multiplied by the probability of survival to n. 0 = n px P 0 n = lx+n /lx = n px . The present value of the insurer’s loss from the portfolio of lx pure endowments is then given by ˜ L = lx+n S n −1 − lx 0 + lx P 0 n −1·S n −1 (3. This term is the difference between the price P 0 n of the bond at time 0 and the discounted amount received by the company at time n. and the (3. the insurer is able to replicate the liabilities. since the value at a future time t of the bonds purchased at time 0 is given by lx n px P t n = lx+n P t n The argument used for determining the price at time 0 of the guaranteed part of the pure endowment can now be repeated at time t for each of the lx+t remaining policy holders. which is exactly the expected number of survivors. in addition. we see that ˜ L = lx+n − lx Sn −1 + lx P0 n − 0 Here.17) determine the fair premium (market price) for the guaranteed payments as well as an investment strategy: that the insurer should invest the entire premium (the market value) in n-bonds.17) ˜ Thus we have obtained that L = 0. as in the classical situation with deterministic interest. the first term is equal to zero exactly if second term is zero if. The first term in Equation (3.3. and the second term represents the present value of the loss from buying at time 0 exactly lx n-bonds at the price P 0 n . we have discounted payments by using the true interest rate (the market interest rate).16). In this way.16) corresponds to the present value of the company’s loss without investments in n-bonds (but investment in the savings account). By rearranging terms in Equation (3. Note that the arguments leading to Equation (3. The fair premium given by Equation (3. the total market value at time t of the .16) where we have assumed that the number of survivors follows the decrement series. the insurer should purchase lx n px = lx+n bonds.

Any increase or decrease of the bond price leads to exactly the same changes in the value of the assets and liabilities. Consider for simplicity the case where the sum insured (one unit) is payable at the end of the year. If it were possible to purchase and sell such contracts at prices which deviated from the market value. The number of deaths in year t predicted at time 0 by the decrement series is given by dx+t = lx+t − lx+t+1 where we have used traditional actuarial notation. Note that the market value of the guaranteed payments does not depend on the company’s choice of investment strategy. so that the company now invests in zero coupon bonds with expiration times t = 1 2 n. This expression can be rewritten as follows: ˜ L= n dx+t−1 − lx t=1 t St −1 n + lx t=1 t P0 t − 0 . at times t = 1 2 n.54 Interest rate theory in insurance guaranteed payments associated with these lx+t pure endowment contracts is given by lx+t n−t px+t P t n = lx+n P t n which shows that the market value of the liabilities is exactly equal to the value of the investments (the assets) at any time t for any future development of the value of the zero coupon bonds. see Møller (2000) and Steffensen (2001). In this case. The company’s loss at time 0 then becomes ˜ L= n dx+t−1 S t t=1 −1 n − lx 0 + lx t=1 t P 0 t −1·S t −1 (3.e. For a more detailed treatment of these aspects. i. we indicate how this hedging argument can be applied for the pricing and hedging of the guaranteed part of a term insurance. it would in fact be possible to generate risk-free gains (arbitrage) by investing in zero coupon bonds.18) which can be interpreted in the same way as for the pure endowment. Hedging a portfolio of term insurances For completeness. even though the market value is derived by means of a hedging argument. Assume more precisely that the company at time 0 invests in lx t t-bonds at the price P 0 t . the argument used for the pure endowment has to be modified slightly. Consider a portfolio of lx term insurances.

that in practice one would work with finitely many different payment times (for example once every day. the policy holder receives the guaranteed amount ba 0 . which would amount to choosing some discretization of Equation (3. and the second term is zero if n 0 = t=1 t−1 1 qx P0 t (3. week or month). . we need a continuous version of (3.19) This fair price differs from the classical formulas in that the usual discount factors have been replaced by zero coupon bond prices. it is no longer possible to ensure that L = 0. of course. i. the payments can occur at any time.e. we mention that if the sum insured is payable immediately upon death and not at the end of each year as suggested by Equation (3.19) becomes 0 = n 0 t px x + t P 0 t dt (3. Accordingly. However. we denote by ba t the amount guaranteed at time t. the first term is zero if t = dx+t−1 = lx t−1 1 qx = t−1 px 1 qx+t−1 which is exactly equal to the probability that a person aged x at time 0 dies in the interval t − 1 t .19). in principle. we can move L arbitrarily close to zero by choosing sufficiently many zero coupon bonds. there are. no matter how many different bonds one buys. this can be obtained by considering small time intervals and noting that. where the policy holder is x years old. Finally. As in the previous chapter. for small h. For example.7 Market values and zero coupon bonds Consider now the main example from Chapter 2 with an endowment insurance with a continuously payable premium . 3.20). in principle.3. We assume in addition that the contract starts at time 0. Another aspect is. Upon survival to n.19). we assume that bonuses are used to increase the amount payable upon survival only. so that ba t ≥ ba 0 . The reason for this phenomenon is that.20) ˜ In this situation. Equation (3. t h qx ≈ t px x+t h Thus.2 Valuation by diversification revisited 55 Here. ˜ infinitely many possible payment times.2. whereas bad is payable immediately upon a death before time n.

56 Interest rate theory in insurance The first order basis Using the notation of Chapter 2. we can determine the premium under the deterministic first order basis r ∗ ∗ via ∗ a∗ = ba 0 n Ex + bad A1∗ xn xn calculated where ∗ n−t Ex+t = exp − n t n t n t r∗ r∗ r∗ d d d ∗ n−t px+t a∗ x+tn−t = A1∗ x+tn−t = exp − exp − s t s t ∗ s−t px+t ds ∗ ∗ s−t px+t x + s ds and where the survival probability under the first order valuation principle is given by ∗ n−t px+t = exp − n t ∗ x+ d Similarly. The technical reserve at time t for the payments guaranteed at time t is given by V ∗ t = ba t ∗ n−t Ex+t ∗ + bad A1∗ x+tn−t − ax+tn−t More generally. we have the following differential equation with side conditions: u V ∗ t u = r∗ u V ∗ t u + V∗ t t = V∗ t V ∗ t n = ba t where R∗ t u is the sum at risk: R∗ t u = bad − V ∗ t u − ∗ x + u R∗ t u (3. we can give prospective expressions for the technical reserve under the first order valuation principle. the technical reserve at time u for the payments guaranteed at time t is given by V ∗ t u = ba t ∗ ad 1∗ n−u Ex+u + b Ax+un−u − a∗ x+un−u For u ≥ t.21) .

e.22) by replacing ba 0 by ba t . From a theoretical point of view. the only difference between the benefits and the premiums is the sign. is given by ba 0 P t n n−t px+t where n−t px+t is the true survival probability.3. the market value at time t of the future premiums is given by n Pt s t s−t px+t ds Note that the market value at time t of the future premiums is calculated in the same way as the market value of the benefits. the situation becomes much more complicated if one includes the policy holder’s possibilities for surrender and for changing the contract into a free policy as indicated in Chapter 2. all formulas there rely on this assumption. i. This quantity is obtained from Equation (3. guaranteed at time 0. In fact. Similarly. Combining the three expressions above. it follows that the market value (the fair price) at time t of the payments upon survival to n. When the actual basis r is deterministic. by repeating the combined diversification and hedging argument of Section 3.23) . However. Section 3.5 shows that the zero coupon bond prices are given by P t s = exp − s r t d (3.2 Valuation by diversification revisited 57 Market value of guaranteed payments In Chapter 2 it was assumed that the mortality intensities and ∗ and the interest rates r and r ∗ were deterministic functions. when r and are both deterministic functions.2. They both involve the zero coupon bond prices at time t instead of the usual discount factors.2. However.22) We denote by V g t the market value at time t for the payments guaranteed at time t. the market value at time t for the part of the contract which pays bad immediately upon a death before time n is given by bad t n Pt s s−t px+t x + s ds where P t s is the price at time t of a zero coupon bond with expiration time s.6. we get the following expression for the market value at time t for payments guaranteed at time 0: V g 0 t = ba 0 P t n + n n−t px+t s−t px+t Pt s t x + s bad − ds (3.

We address questions like: What is the term structure of interest rates? What is a forward rate. A natural question is therefore: does this solve the problem of determining market values of the guaranteed payments from a life insurance contract completely? A quick answer is: yes! However. We can only use this simple formula when the market interest rate is assumed to be deterministic or even constant. and what is the difference between forward rates and the market interest rate? What is credit risk? What is the role of these concepts in the calculation of market values in life and pension insurance? We start by analyzing zero coupon bonds. 3.25) = ba t n−u Ex+u + b A1 n−u − x+u which corresponds to the formulas derived in Chapter 2. we recall some fundamental concepts related to bond markets.5 below we discuss connections to coupon bonds. this leads to the question of whether it is possible to construct versions of Thiele’s differential equation for the market value of the guaranteed payments. In Section 3. One aspect is that zero coupon bonds are not determined from the true market interest rate r via Equation (3.23) if the market interest rate is stochastic.24) (3.3 Zero coupon bonds and interest rate theory In this section. we insist on continuing the discussion here for various reasons. Here. .58 Interest rate theory in insurance If we insert this expression into Equation (3. the market value at time u for the payments guaranteed at time t can be written as follows: V g t u = ba t exp − + n u n r u s d d n−u px+u exp − r u ad s−u px+u x + s bad − ax+u n−u ds (3. In particular. This fact requires a more detailed treatment of the relation between the prices of these more general bonds and zero coupon bonds. Another aspect is that the majority of bonds on most bond markets are more complicated than zero coupon bonds. T and T are fixed (deterministic) finite times. the concept of forward rates proves to be useful. What’s next? The above calculations indicate that the market value of the guaranteed payments can be calculated by replacing the usual discount factors with zero coupon bond prices. In the following. which can be viewed as the basic building blocks of the bond market.22).

see Bielecki and Rutkowski (2001).3. but that the intensity depends on the price of the zero coupon bond at time t. Definition 3. that the issuers of the bonds are not likely to go bankrupt. This quantity is exactly the forward rate.1 Yield curves Definition 3.e. Often the contracts defined above are called default free zero coupon bonds.2 Forward rates Now consider times 0 ≤ t ≤ T ≤ T .3 Zero coupon bonds and interest rate theory 59 Definition 3. Note that this (constant) interest rate R t T applies for any time u ∈ t T .1 A zero coupon bond with maturity date T (also called a T bond) is a contract which pays one unit at time T . The price at time t ∈ 0 T is denoted by P t T . i.26) can be interpreted as a discount factor obtained by using the (constant) interest rate R t T during the interval t T . As a consequence.3. since the bonds may be worthless if the issuer goes bankrupt. This is in contrast to so-called defaultable bonds.26) so that Equation (3. 3. Throughout. We cannot . In the following.2 The continuously compounded zero coupon yield (or the continuously compounded spot rate) R t T for the period t T is defined by R t T =− 1 log P t T T −t It follows directly from this definition that the price of the zero coupon bond can be expressed in terms of the continuously compounded yield as follows: P t T = exp −R t T T −t (3. This risk is also known as credit risk.3. This serves to underline that the bonds cannot (or are very unlikely to) default. bonds issued by such parties typically give a higher return to the holders. we focus on default free bonds. we take P t t = 1. We are interested in finding a deterministic rate at time t (today) for a future investment made at time T and terminated at time T . which are issued by companies (or countries) which are less credit-worthy.3 The term structure of interest rates at time t (or the zero coupon yield curve) is given by the mapping h → R t t+h 3. Lando (2004) and Schönbucher (2003).

P t T /P t T . use P t T to buy P t T /P t T units of T -bonds. provided that the mapping T → P t T is continuously differentiable with respect to the maturity date T . In our subsequent analysis in Section 3. i.4 The continuously compounded forward rate at time t for the period T T is defined by log P t T − log P t T (3. in general. known at time t.60 Interest rate theory in insurance use the yield R T T since this is defined in terms of P T T . is not known at time t.5 of market values in life and pension insurance. Instead. Equation (3.28) This can be given the following interpretation: • • • • at time t. So. at time T pay one unit on the T -bond (pay back the loan). which. A result of these transactions is that we have to pay (or invest) one unit at time T .3. Definition 3. Thus.27) shows that this is simply the derivative of − log P t T with respect to T . we can state the following definition. from Equation (3.27) T −T A better understanding of the concept of forward rates can be obtained by noting that the ratio of the price at time t for a T -bond and a T -bond is given by f t T T =− Pt T Pt T = exp −f t T T T −T (3. still at time t. the limit of f t T T as T T. sell one T -bond and receive P t T (this corresponds to a loan). Furthermore. at time T receive P t T /P t T from the T -bond. it is useful to work with the instantaneous forward rate at time t for a given future time T . At time T we receive the result of this investment. one introduces the concept of forward rates. . which should be used for discounting future payments from time T to time T . by trading zero coupon bonds at time t we are. since the T -bonds are financed by selling a T -bond. in a sense. The two transactions at time t are chosen such that the result at time t is exactly 0.28) we obtain that Pt T = exp f t T T T − T Pt T This shows that f t T T can be interpreted as the constant interest rate.e. able to fix the future return (or interest) for an amount to be invested at a future time T .

30) Note. We will not go into a treatment of the mathematical aspects associated with these different methods. there are several possibilities for specifying models for a bond market: • specify all P t T for 0 ≤ t ≤ T ≤ T ∗ . where T ∗ is some maximum time point. for 0 ≤ t ≤ T ≤ T ∗ .5 The instantaneous forward rate at time t for time T is defined by f t T =− T log P t T (3.5. To obtain this result. we mention that one has to be careful when using the first two approaches in order to avoid arbitrage possibilities.30) and the instantaneous short rate r t via Definition 3. which basically means that the instantaneous short rate is deterministic. • specify the development of r t . that we cannot in general conclude that the zero coupon T bond prize is of the form P t T = exp − t r u du . It follows immediately from the definition above that T ft t d = − log P t T + log P t t Since P t t = 1. • specify all f t T for 0 ≤ t ≤ T ≤ T ∗ and derive P t T via Equation (3. However. More precisely. it is necessary to assume that future bond prices are known at time t.2. we see that the value at time t of the zero coupon bond is given by P t T = exp − T ft t d (3.3 Zero coupon bonds and interest rate theory 61 Definition 3. additional assumptions are required.5. Thus. since it is not clear how r determines the zero coupon bond prices. The last approach also requires some additional work.29) The instantaneous short rate at time t is r t = f t t .3. We show in the following how forward rates appear in a natural way in Thiele’s differential equation for the market value of the guaranteed payments from a standard life insurance contract.30) shows that the zero coupon bond prices can in fact be derived directly from the instantaneous forward rates. . This situation was considered in Section 3. Equation (3. however.

such as zero coupon bond.28) with T = t. spot rate and forward rate. in our treatment of market values in life and pension insurance in the present chapter. one can also introduce simple yields (or simple spot rates) and simple forward rates. Thus.3. .62 Interest rate theory in insurance 3. the forward rate at time t for t T coincides with the spot rate for t T.30). and the reader may wonder if all this is really necessary and how all these quantities are related. They reappear in our discussion of swap rates in Chapter 7.31) which shows that the zero coupon yield for t T can be interpreted as the average of the instantaneous forward rates. (2) Combining Equations (3. this leads to the interest PL t − s for the time period s t .4 Simple rates As an alternative to the continuously compounded rates introduced above. however. we see that 1 T −t T ft t d =R t T (3.3. the deposit on the savings account would increase from Pers at time s to Pert at time t. For completeness. With continuously compounding interest under constant interest rate r. (1) From Equation (3. which does not play a further role.26) and (3. If we deposit an amount P at time 0 in an account with simple rate L (per time unit). 3. we give a short discussion of this concept.3 Relations between forward rates and spot rates We have now introduced several quantities.e. Simple rates differ from the usual principle of compounding rates in the sense that past interest is not included in the calculation of the interest for a given period. the interest credited during the interval s t is given by P ert − ers = Pers er t−s − 1 This can now be formalized via the following definition. Some immediate consequences of the above definitions are as follows. yield. it follows that f t t T =R t T i.

Under the principle of simple interest.3.5 Market values and forward rates In this section we derive a version of Thiele’s differential equation for the market values of the guaranteed payments that involves forward rates.3. If we insert this expression into Equation (3. 3.3. we immediately obtain the following expression: V g t u = ba t exp − + n u n fu u s d d n−u px+u exp − fu u s−u px+u x + s bad − ds (3.30).2 we showed how the instantaneous forward rates f t at time t are related to the price of a zero coupon bond at time t. In Section 3. see Equation (3. the amount P t T L t T T −t is exactly the interest which accrues in the interval t T in connection with the investment of P t T under a constant simple interest L t T . In this situation.25) cannot be applied directly. we can derive an alternative expression for the market value .6 The simple yield (or the simple spot rate or even the LIBOR spot rate) for t T is given by L t T =− P t T −1 T −t P t T and the simple forward rate (or the LIBOR forward rate) for T T at time t is given by L t T T =− P t T −P t T T −T P t T It follows from the definition of the simple spot rate that L t T T −t P t T = 1−P t T which has the following interpretation: 1 − P t T is the return or gain during the interval t T from buying the T -bond at time t at the price P t T and cashing the amount 1 at time T . This auxiliary quantity is of importance in situations with stochastic interest.25) for the market value at time u ≥ t for the payments guaranteed at time t.3 Zero coupon bonds and interest rate theory 63 Definition 3.32) The situation t = u gives the market value V g t at time t of the payments guaranteed at time t. where Equation (3.

The most simple situation is the case where − x + u Rg t u (3. since V g t u is a rather complicated function of u which moreover depends on the interest rate at time u. for example. It is not possible to derive a differential equation for V g t u by applying Equation (3. Schwarz (1989).34) on the interval t n with the terminal condition corresponding to the payment upon survival to n. the market value V g t can be calculated by solving Equation (3. Given the sum insured ba t guaranteed at time t and the forward rates f t u at time t. see. The individual bonus potential The forward rates can also be used for deriving an expression for the so-called individual bonus potential V ib . For a description of the methods for solving differential equations numerically. However. given by Rg t u = bad − V g t u This differential equation provides an alternative method for calculating Equation (3. Hence we obtain the following differential equation for u ≥ t with side conditions: u Vg t u = f t u Vg t u + Vg t t = Vg t V g t n = ba t where Rg is the sum at risk. Instead. which are known at time u.64 Interest rate theory in insurance which is similar to the classical formulas.33) as follows. calculated by applying the instantaneous forward rates at time t. the main advantage of this quantity is that it is a very simple function of u. we introduce the additional auxiliary function V g t u = ba t exp − + n u n ft u s d d n−u px+u exp − ft u s−u px+u x + s bad − ds (3. but where the stochastic interest rate is replaced by the instantaneous forward rates at time u. Thus.33) which may be interpreted as a value at time u for the payments guaranteed at time t. which also opens up the possibility of deriving a differential equation for the individual bonus potential in a model with stochastic interest rates.34) . guaranteed at time t. We give here a version of Thiele’s differential equation. V g t u is not really a market value. which can be differentiated directly.32).

The estimated mortality intensities are clearly ordered and reveal a decreasing trend from 1970 to 2003. see Chapter 2. the individual bonus potential is given by V ib t = V ∗ t − V g t By applying the differential equations.29)). Figures 3.37) The actual parameters used are listed in Table 3.34). 3. We consider an insurance contract. we illustrate the role of the bonus potentials via a numerical example. The expression in Equation (3.21) and (3. The estimated mortality intensities can be found in Figures 3. these figures are also taken from Dahl and Møller (2006). together with the estimates for 1970 and 1990.1. we find that V ib t = where c t s = f t s − r∗ s V ∗ t s + ∗ n t exp − s ft t + x+ d c t s ds (3. where V t is the total reserve for the contract. This serves to underline the importance of the forward rates in the situation where the interest rate is stochastic. .4 show the expected lifetimes for age 30 and age 65. together with the terminal condition V ∗ t n − V g t n = ba t − ba t = 0. This decrease in the mortality intensity implies that the expected lifetimes have increased during the same period.36) Equations (3.4 A numerical example 65 V t ≥ V ∗ t ≥ V g t .28) and (2. since the safety loadings c t s for the payments guaranteed at time t now involve the forward rates at time t.1 and 3. In this case.2.3 and 3.4 A numerical example In this section. Equations (3.35) x+s − x + s R∗ t s (3.28) in Chapter 2: the individual bonus potential is the market value of the safety loadings on the guaranteed payments.35) has the same interpretation as Equation (2. where the benefits have been calculated using a first order mortality intensity of Gompertz–Makeham form as follows: x+t = + cx+t (3. This table also contains Gompertz–Makeham estimates for the Danish population taken from Dahl and Møller (2006) for males and females for 1980 and 2003.35) and (3. in that the market interest rate has been replaced by the forward rates.36) differ from the corresponding expressions in Chapter 2 (see Equations (2.3.

Males c First order 1980 2003 0 0005 0 000233 0 000134 0 00007586 0 0000658 0 0000353 1 09144 1 0959 1 1020 0 0005 0 000220 0 000080 0 000053456 0 0000197 0 0000163 Females c 1 09144 1 1063 1 1074 0. Estimated mortality intensities for females.1. the expected lifetime in 2003 is about 75.5 years for females aged 30 from 1980 to 2003.10 0. dotted lines to 1990. In contrast.00 30 40 50 60 70 80 Figure 3. whereas the estimates for c decrease. and dot-dashed lines to 2003.2.5 for 30-year-old females. dotted lines to 1990.1.2 shows the expected lifetimes . Estimated mortality intensities for males. 0.10 0. and dot-dashed lines to 2003. dashed lines correspond to 1980.00 30 40 50 60 70 80 Figure 3. Figure 3. Solid lines are 1970 estimates. Table 3.66 Interest rate theory in insurance Table 3. A closer study of the underlying Gompertz–Makeham parameters c indicates that has decreased during the period from 1960 to 2003 for both males and females. decreases and c increases in the last part of the period considered (from 1990 to 2003). Using this method. The first order mortality and estimated Gompertz–Makeham parameters for 1980 and 2003.5 years for males and 1.20 0.3 shows that the remaining lifetimes have increased by approximately 2. Solid lines are 1970 estimates. dashed lines correspond to 1980.3 years for 30-year-old males and 79. The estimates for increase from 1960 to 1990.

9 81. In this example.9 81.8 82. With the 2003 estimate. 2003 Est. Development of the expected lifetime of 30-year-old females (dotted line) and 30-year-old males (solid line) from 1960 to 2003.9 93.e. 1980 Est.4 A numerical example 67 Table 3.3.1 78.8 93. respectively. for future reductions in the mortality.1 78.5 93.7 and 2. 65 and 90.2 93.5% and 1%. Age First order Est. 2003 (longevity 0. we see that the expected lifetimes calculated under the first order basis are smaller than the expected lifetimes calculated under the 2003 estimate.8 94.9 73. 84 80 76 1960 1970 1980 1990 2000 Figure 3.4 94.2 83. calculated for the first order basis and the estimated mortalities for age 30.0 75.3 94. For males and females aged 30 or 65. we have included simple corrections for longevity.2 years.4 80.1 83.3 79.3 93. 2003 Males 30 65 90 74.4 Females 30 65 90 78 74 70 1960 1970 1980 1990 2000 Figure 3.1 78.8 80.4 82.4. i. Estimates are based on the last five years of data available at the given years. Development of the expected lifetime of 65-year-old females (dotted line) and 65-year-old males (solid line) from 1960 to 2003.5%) (longevity 1%) 77.5 94. by assuming that the mortality in all ages decreases each year by 0. the difference is between 1.3 Est.7 77.4 84.4 94.3. If we include a correction for future improvements in the . Expected lifetimes with the first order basis and the market-value bases for 1980 and 2003. moreover.5 84.2.8 80.

can be found in Table 3. These numbers.3 for maturity t = 1 2 30. We have used this yield curve for the calculation of all market values below.095 1960 1970 1980 (c) 1990 2000 Figure 3.68 Interest rate theory in insurance 4 × 10–4 1 × 10–4 1960 1970 1980 (a) 6 × 10–5 1990 2000 2 × 10–5 1960 1. Development of the Gompertz–Makeham parameters: (a) . The expected present values used in the example can be found in Table 3.5. which have essentially been adjusted for tax by multiplying the observed yield curve by 0. In the example. in particular at low ages. we use the value for maturity 30. we see that this difference is now between 2. mortality. Using a yearly decline of 1%. for t ≥ 30.115 1970 1980 (b) 1990 2000 1. (b) .6 years. Estimates are based on the last five years of data available at the given years. we have determined market values by using the above 2003 mortality intensity with a yearly 1% correction for future mortality improvements.7 and 6.105 1. we see that this difference increases considerably.4 for first order bases with (approximate) interest . (c) c for females (dotted lines) and males (solid lines) from 1960 to 2003. we have applied a zero coupon yield curve from 31 December 2005 for discounting the payments. In addition.85.

this leads to .5% and 1.03737 0. Premiums are paid continuously at a fixed rate as long as the policy holders are alive and stop at the retirement age of 65.5 for the three contracts.03277 0.02973 0. At the time of signing these contracts at age 30.03489 0.03573 0.02867 0.5%.03084 0. These factors are used for calculating the market value V f of the guaranteed free policy benefits.5 also includes the so-called free policy factors introduced in Section 2. The relevant market values can be found in the lower part of Table 3.02422 0. 2. such that the technical reserves start at zero at time 0.5%. see Kalashnikov and Norberg (2003).03750 0.02632 0.3.03746 0.02901 0.3.03114 0.5% and 1.03706 0.02843 0.3 and the estimated mortality from 2003 with a yearly longevity correction of 1%.03615 Yield after tax 0. For each of these three contracts. respectively.5%.03090 0.5 for the three contracts with different technical interest rates.03771 Yield after tax 0 03100 0 03120 0 03139 0 03159 0 03178 0 03182 0 03185 0 03189 0 03193 0 03196 0 03200 0 03203 0 03207 0 03211 0 03214 rates 4.03767 0.5%. We have then calculated the technical reserve at various ages by rolling forward the policy holders’ accounts with the first order assumptions. respectively.4 A numerical example 69 Table 3.03045 0. we see that the market values of the guaranteed payments are negative for the two contracts with low technical interest rates (2.02653 0. we have determined the equivalence premium at age 30.02991 0.03203 0.03638 0.02937 0.03763 0. (For a study of the dependence of the technical reserve on the interest rate and other parameters.03009 0.5%). Zero coupon yields from 31 December 2005 before and after tax.03365 0.03729 0.02729 0.02548 0. We consider now three contracts signed at age 30 with first order interest rate 4.7.02792 0. We set the life annuity payment rate to unity and sum paid at death to five.02627 0.03758 0.03531 0. Maturity 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Yield 0.5% and 1. This table shows (not surprisingly) that the premiums and technical reserves increase if the technical interest decreases.) Table 3. This table also contains the expected present value calculated by using the zero coupon yield curve of Table 3. We see that the factors decrease when the technical interest is decreased.03660 0.02694 0.03404 0.03754 0.03081 Maturity 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Yield 0.03683 0. 2.03733 0.03741 0. The fair premiums can be found in Table 3.03447 0.1.03162 0.

The life annuity is assumed to start at age 65. pure endowment and the life annuity. We see that the contract with the high technical interest does not lead to any bonus potentials.5%.5%.4. the market value is greater than zero at time 0 for the contract with technical interest 4. term insurance. small bonus potentials arise because the first order mortality is lower than the estimated mortality at high ages. we see that the market values of the guaranteed free policy benefits are typically higher than the market values for the guaranteed payments. For the contracts with technical interest 2. Premiums are paid continuously and are assumed to stop at age 65. In contrast.5% Market values. which implies that the company needs to set aside additional capital for this contract. (At very high ages.5%.5% 30 40 50 65 90 30 40 50 65 90 30 40 50 65 90 30 40 50 65 90 17 011 14 273 10 281 0 000 0 000 22 069 17 406 11 679 0 000 0 000 25 519 19 384 12 492 0 000 0 000 20 709 16 804 11 595 0 000 0 000 0 086 0 110 0 122 0 000 0 000 0 130 0 145 0 143 0 000 0 000 0 161 0 168 0 155 0 000 0 000 0 068 0 090 0 103 0 000 0 000 0 165 0 262 0 425 1 000 0 000 0 323 0 423 0 568 1 000 0 000 0 455 0 540 0 657 1 000 0 000 0 278 0 384 0 546 1 000 0 000 1 689 2 681 4 356 10 239 3 277 3 880 5 080 6 810 11 997 3 460 5 952 7 069 8 595 13 080 3 560 3 753 4 994 6 798 12 188 3 200 First order interest 2. 1% longevity bonus potentials at the time of signing the contract.) For the two contracts with technical interest 1.70 Interest rate theory in insurance Table 3. Expected present values (for males) under the first order valuation basis and the market-value basis for the premiums.5% First order interest 1. however.6.5%.5% and 2. this is not the case. The individual bonus potentials for this example are listed finally in Table 3. For the contract with interest rate 4. the bonus potentials on the future premiums . where the term insurance and pure endowment expire. Age Premiums Term insurance Pure endowment Life annuity First order interest 4.5% and 1.

we give a framework that can be used for basically any bond with predetermined payments. Premiums Interest Age 30 40 50 65 90 Age 30 40 50 65 90 1 511 3 347 5 868 12 188 3 200 0 000 1 451 3 685 10 239 3 277 0 125 4 50% 0 205 2 50% 0 265 1 50% 4 50% 2 50% 1 50% Technical reserves 0 000 2 235 5 128 11 997 3 460 Market values. the bonus potentials on the free policy start at zero and increase with the premiums paid.5 Bonds. The market values of guaranteed payments (GP) and market values of guaranteed free policy payments are given in the lower part. interest and duration The majority of bonds traded in most bond markets are more general than zero coupon bonds. Typical examples are annuity bonds and bullet bonds. with continuously paid premiums. GP −0 157 1 994 4 934 12 188 3 200 −1 393 0 991 4 242 12 188 3 200 0 000 2 776 6 062 13 080 3 560 0 000 0 449 0 742 1 000 1 000 Free policy factors 0 000 0 385 0 682 1 000 1 000 0 000 0 351 0 647 1 000 1 000 Market values. respectively (upper part).5% and 1.5 Bonds. which shows how the prices of these more general bonds are related to the prices of zero coupon bonds. However.1 A general bond Consider a bond issued at time 0 . technical reserves and free policy factors for the three policies with first order interest 4. The value of such a bond at time t ≥ 0 cn (after .5. say.5. which specifies payments c1 at given times 1 < · · · < n . In contrast. 2. at some point. 3.5%. Contract: term insurance of five units and life annuity of one unit. Fair premiums (equivalence premiums fixed at age 30).5%.3. Here. they start to decrease again and eventually vanish as the age increases. free policy 0 000 2 445 5 426 12 188 3 200 0 000 2 095 4 984 12 188 3 200 0 000 1 910 4 731 12 188 3 200 decrease as the age approaches the age of retirement. interest and duration 71 Table 3. 3.

Consider the situation where ck = L for k = 1 n − 1 and cn = L n − n−1 k − k−1 K K +K . free policy 0 000 0 000 0 000 0 000 0 077 0 000 0 140 0 144 0 000 0 260 0 000 0 866 1 331 0 892 0 360 Bonus potential. premiums 0 000 0 000 0 000 0 000 0 000 0 157 0 100 0 050 0 000 0 000 1 393 0 919 0 489 0 000 0 000 possible payments at t) is given by Pt = i i >t Pt i ci (3. the additional reserve needed and the two individual bonus potentials (BP): on premiums and on the free policy. This can be seen by noting that payment streams of the form ck cn at times k < · · · < n can be generated by buying exactly ck k -bonds. a simple arbitrage argument shows that this is the only price that does not lead to the possibility of generating risk-free gains in a market where it is possible to buy and sell the bond in Equation (3.38) as well as zero coupon bonds with maturities 1 n .72 Interest rate theory in insurance Table 3.38) where P t i is the price at time t of a zero coupon bond with maturity i . Again.7 (Annuity bond) Taking ck = c for k = 1 bond. ck+1 k+1 -bonds.50% 1.50% 2. Interest Age 30 40 50 65 90 Age 30 40 50 65 90 1 511 3 347 5 868 12 188 3 277 4 50% 2 50% Total liability 0 000 2 235 5 128 12 188 3 460 0 000 2 776 6 062 13 080 3 560 1 50% 4. n gives an annuity Example 3.6. The total liability.8 (Bullet bond) Fix some simple rate L and some amount K (the principal).50% Additional reserve included 1 511 1 896 2 183 1 949 0 000 0 000 0 000 0 000 0 191 0 000 0 000 0 000 0 000 0 000 0 000 Bonus potential. Example 3. etc.

then the mapping y t → F t y t is strictly decreasing. the yield to maturity y t is the constant rate which ensures that the discounted value of all future payments corresponds to the value at time t of the bond. how will a change in the yield to maturity affect the bond price? It can be relevant to investigate. . Definition 3. This can be made more precise via the following definition. y t is the constant rate of return that the holder receives for the payments during t n .9 The yield to maturity at time t of the bond in Equation (3. which can be interpreted as the weighted average of the time to the payments ci ≥ 0 weighted by the factors ci e−y t i −t . Thus.3. These factors represent the values at time t of the payments ci discounted by using the yield y t . we introduced the zero coupon yield R t T . At time n .5. Here. interest and duration 73 With this bond. the principal K is paid back together with interest for the interval n−1 n . 3. Similarly. for example.3.1. The precise definition of the concept of duration is given here. The answers to these questions are closely related to the so-called duration of the bond.38) is the solution y t to the following equation: Pt = i i >t e−y t i −t ci = F t y t (3. and hence there exists a unique non-negative solution to Equation (3.2 Yield and duration In Section 3.39).39) Note that if all ci are non-negative and ci > 0 for some i with i > t. which was defined as the constant intensity for t T which allows the zero coupon bond with maturity T to be expressed as a traditional discounting factor of the following form: P t T = exp −R t T T − t A natural generalization of this concept can be given for more general bonds. one might be interested in a comparison of the sensitivity of different bonds. A natural question is now the following: how sensitive is the price P t of the bond to changes in the yield to maturity? Or. how sensitive the value of the company’s total portfolio of bonds is to changes in the term structure. the holder receives at time k the simple interest L k − k−1 on the principal K for the interval k−1 k . more precisely.5 Bonds.

(We use this notation in order to emphasize the dynamic nature of this problem.40) We see from this definition that the duration at time t for a zero coupon bond with maturity is exactly equal to the time to maturity − t. 3. where J t = j∈ 1 n j >t i. Hull (2005) and references therein. what can be said about the zero coupon yield curve and the forward rates? It is important to note that so far we have not introduced any model to describe how the price P t T of a given zero coupon bond evolves over time.74 Interest rate theory in insurance Definition 3. for i ∈ J t . If we think of P t in Equation (3. we could simply consider one fixed time.39) as a function of the yield to maturity y t . say. t = 0. 3.6. for example.e. see.) We can now derive the zero . We consider the most simple situation where we observe the prices at time t of some zero coupon bonds and address the problem of deriving the forward rate curve. we simply think of the prices as given and try to extract as much information from the observations as possible without introducing complicated models.6 On the estimation of forward rates In practical situations. one is often confronted with the following problem. This will be done later.38) with yield to maturity y t is defined by Dt = i i >t i −t e−y t i −t ci Pt (3.39) with respect to y t . the following: d P t =− dy t i i −t i >t e−y t i −t ci = −P t D t (3. and work with a fixed set I of observed bond prices with maturities 1 n . we obtain. At this point. Given that we have observed prices of a number of bonds. of bonds that have not expired at time t. by differentiating Equation (3. Alternatively.10 The duration D t at time t of the bond in Equation (3.41) For more details on duration.1 Observing prices of zero coupon bonds We fix times 1 < · · · < n and assume that we have observed at time t the prices P t i of zero coupon bonds with maturities i .

6 On the estimation of forward rates 75 coupon yield for the periods t shows that Rt i i . see Svensson (1995).2. one can fix some parameterized family of possible instantaneous forward rate curves. by use of Definition 3. we review the so-called Nelson–Siegel parameterization for the forward rates. which are the pairs of time to maturity and zero coupon yield for the corresponding period. Nelson and Siegel (1987) advocated the following parametric family for the forward rates: f0 = 0+ 1e − / + 2 e− / (3. we fix t = 0.31). h → f t t+h ∈ i and fit this curve to the estimated zero coupon yields R t using Equation (3. we are also interested in (an estimate for) the entire zero coupon yield curve h → R t t + h at time t or the curve for the instantaneous forward rates h → f t t + h .42) .3. Definition 3. the problem consists in finding the curve R t t + h which provides the best fit to the observed quantities R t i . Alternatively. by using some subjective criterion.4 determines the forward rates for the periods i−1 i . given by ft i−1 i =− log P t i − log P t i − i−1 i−1 However. which shows that Rt i . by = 1 i −t i ft u t du Thus. i ∈ J t . suggested by Nelson and Siegel (1987). This is discussed in more detail below.6. which for i ∈ J t =− 1 log P t i −t i One can visualize this by plotting the points i − t R t i . One typical approach to the first problem is to apply some numerical technique to fit the best smooth curve to the observed points. 3. i ∈ J t . for an extension and an application to the Swedish Treasury bills and coupon bonds.2 The Nelson–Siegel parameterization In this section. For simplicity. i ∈ J t . i ∈ J t .

8 forward rates 0. This special property can be exploited to yield a very simple approach to fitting the model to observed prices using 1. and the exponentially decreasing line affects “short term” forward rates. the parameter is determining how “short term.43) The different components determining the forward rates are depicted in Figure 3. . We now discuss a simple way of fitting Equation (3.” and R0 where Yi = 1 1 − e− i/ i = 1+ 2 1 − e− i/ 1 + i/ 3e − i/ = Yi tr (3. the part with 1 is determining the short term forward rate.2 0.6.6.44) 1 i/ tr e− i/ Thus. where tr means “transposed.0 0. we first change the parameterization slightly and introduce the parameter = 1 2 3 tr . we find that the zero coupon yield is given by R0 = 1 0 f 0 t dt = 0+ 1+ 2 1 − e− / 1 − / 2e − / (3. R 0 i is linear in . can be interpreted as a long term component. and the 2 -component is affecting the medium term forward rate. 0 . The first part. In order to simplify notation.6 0. Finally.” “medium term” and “long term” should be defined. the dashed line affects “medium term” rates. for given .0 0 2 4 maturity 6 8 10 Figure 3. Integrating Equation (3. Nelson–Siegel parameterization: components determining the forward rates.4 0.76 Interest rate theory in insurance which is parameterized by = 0 1 2 and .43) to some observed zero coupon yields for the periods 0 1 0 n . The horizontal line gives the level for “long term” forward rates.42) with respect to the maturity .

e. positive definite weight matrix. for fixed .44) to the observed prices is now to minimize over .47) into Equation (3. i. Norberg (2000). Y tr A Y ˆ −R = 0 (3. for example. for example.45) is minimized for solving / h = 0. ˆ Assume that we have observed zero coupon yields R 0 i . see Norberg (2000). i. For example. min h For more details on this approach applied for the estimation of mortality intensities.6 On the estimation of forward rates 77 the generalized least squares method. .) We then introduce the vector ˆ ˆ ˆ of observed yields R = R 0 1 R 0 n tr and the n × 3 -matrix Y given by ⎞ ⎛ 1 tr ⎞ ⎛ 1 1 − e− 1 / 1/ e− 1 / Y 1 ⎟ ⎜ ⎟ ⎜ ⎟ Y =⎝ ⎠=⎜ ⎠ ⎝ 1 n tr − n/ − n/ Y 1 1−e e / n One possible way of fitting Equation (3. (For example. the quadratic function h = 1 Y 2 ˆ − R tr A Y ˆ −R (3. for i = 1 n. one could choose A equal to the n × n -identity matrix. Alternatively.46) which has the following solution: = Y tr AY −1 Y tr ˆ AR (3.3. these could be derived from zero coupon bond prices or from some more general bonds as described above. one could use trading volume as a measure or choose to give more weight to yields from specific maturities.45) where A is some symmetric.47) The optimal choice of can then be found by inserting Equation (3. for example. Equation (3.45) and minimizing h over . also be used to fit a Gompertz–Makeham mortality intensity to occurrence-exposure rates.e. see. This approach can. one could specify A in such a way that the estimated curve is forced to be close to some specific yields. this method is also applied by Cairns (2004) for the estimation of the yield curve.

it can be shown how these instruments should be priced (within a given model) in order to avoid the possibility of generating risk-free gains. where the interest rate for the last period can attain two different values. How can this be exploited to make a risk-free gain? Well. stocks and even savings accounts. Some fundamental results from probability theory that are crucial for a more systematic introduction to arbitrage-free pricing are reviewed in Section A.7.1) S 0 T ≥ S 1 T and that P S 0 T > S 1 T > 0.1 describes the market of traded assets. The theory presented in this section is particularly useful for explaining how so-called derivatives can be priced.11 (An arbitrage) As an example of a risk-less gain or an arbitrage. By the above assumptions. At time 0 buy one unit of asset number 0 and sell one unit of asset number 1. This section is organized as follows. consider the situation where there exist two traded assets S 0 and S 1 whose prices at time 0 coincide. we have P V ≥ 0 = 1 and P V > 0 > 0.2 we consider a simple two-period example with two assets.9 how these results can be applied when calculating market values in life and pension insurance. but it will have at least the same value at time T . In Section 3. Example 3. At . it seems intuitively reasonable that there is something wrong with this model. Assume. Section A. In addition. Section 3.7.78 Interest rate theory in insurance 3. Readers not familiar with concepts such as filtrations. More precisely. Important examples for life insurance companies are options. we see in Section 3. that is S 0 0 = S 1 0 . we assume that we can buy and sell the assets at the given prices. stochastic processes and martingales may find it helpful to consult this appendix for more details. that we know that with probability 1 (almost surely. (Thus. furthermore. see the Appendix. which has the following consequences. and hence we are not dealing with transaction costs. swaptions and interest guarantee options. The term asset is used for basically any traded securities. such as bonds. since asset number 0 seems to be to cheap: it has the same price as asset number 1 at time 0.7 Arbitrage-free pricing in discrete time What is arbitrage-free pricing? What is hedging? What is a martingale measure? How can these concepts be used for the calculation of market values in life and pension insurance? This section provides an introduction to the principle of no arbitrage.) At time T this leads to the gain V = S 0 T − S 1 T ≥ 0. This observation can be exploited in the following way. The main idea is that prices of traded assets should be determined in such a way that no risk-less gains arise from trading these assets.1 of the Appendix.

(Mathematically this means that S t is t -measurable. a filtration is an increasing sequence of -algebras. . however. If they existed.5. (It is not difficult to generalize the model to the case of d traded assets. 3. respectively.11) should exist. all investors would be interested in following these strategies. The price processes and the amount of information available.3.) Here we depart from the previous assumption of deterministic prices as in Section 3. Thus.7. We assume that the amount of information is non-decreasing over time (we do not forget).) The price process S i is said to be adapted to the filtration F if the price of asset i at time t is part of the information t available at i time t. whose prices at time t are given by S 0 t and S 1 t . The main assertion in arbitrage-free pricing is that no arbitrage possibilities (like the one in Example 3. any predictable process is adapted. this would only serve to complicate notation and the presentation in general.2. we introduce a so-called filtration F = t t∈ 0 1 T . Basically. S i = S i t t∈ 0 1 T .7.1 Traded assets and information We consider a financial market consisting of two traded assets. time: 0 1 t−1 t T prices: information: Si(0) (0) Si(1) (1) Si(t − 1) (t − 1) Si(t) (t) Si(T ) (T ) Figure 3. and allow for randomness so that future prices are not known in general. More precisely.) The model is illustrated in Figure 3. A stochastic process is called predictable if its value at time t is already known at time t − 1. this means that we have received a lottery ticket for free.7. t is the information available at time t. the prices at time u are described by the random variables S 0 u S 1 u . which are not observed before time u. so that t ⊆ u for any t < u ≤ T . The family of prices for asset i in discrete time. (Mathematically. is an example of a stochastic process. Here. To keep track of what is known at time t. and this would imply that there would only be buyers for the cheap asset (asset 0 in the example) and sellers for the expensive asset (asset 1 in the example).7 Arbitrage-free pricing in discrete time 79 time T we receive V ≥ 0 without having paid anything at time 0.

We assume that trading takes place at times t = 0 1 2. For example. (For a treatment of such models. defined by X t = S 1 t /S 0 t and X 0 t = S 0 t /S 0 t = 1 3.8. we consider the question of finding the value of the bond at time 1. we consider a simple example with two periods. The value of the savings account at time 0 is given by unity. One can now ask questions such as: is is possible to find the price of a zero coupon bond with maturity T = 2 from these assumptions? First. one can think of S 1 as the value of a zero coupon bond. independently of the value of the interest.8. see. We assume. Assume that the interest i 1 for the savings account for the first period is known at time 0.e.7.e. where it is possible to deposit money on the savings account and receive an interest of 6% at the end of the period. If interest has increased from 5% to 6%. Then.) An illustration of this model can be found in Figure 3. As in the previous sections. where the interest for the first period is taken to be i 1 = 5% and where the interest for the second period is either i 2 = 4% or i 2 = 6%. i. Let p = P i 2 = 0 04 = 1 − P i 2 = 0 06 and assume that 0 < p < 1. S 0 0 = 1.2 A two-period example Before we continue our discussion of arbitrage-free pricing. that the future interest can take two different values only. there is a positive possibility for both events. i t t∈ 1 T is also a stochastic process. for example. we apply the process S 0 as a discounting factor (also called a numeraire) and we also introduce the discounted price processes X and X 0 . we recall that P 2 2 = 1. we are in the upper part of Figure 3. Similarly. i. The usual arbitrage argument now shows that . whereas the interest i 2 for the second period is unknown at time 0 and is only revealed at time 1. Jarrow (1996).80 Interest rate theory in insurance We assume that S 0 describes the development of a savings account with periodic interest i t for the period t − 1 t . S 0 1 = 1 + i 1 = er 1 and S 0 2 = 1 + i 1 1 + i 2 = er 1 +r 2 . in addition. The example p = 1/2 is the case where both events are equally likely. so that S0 t = 1 + i 1 · · · 1 + i t Typically.

092 Figure 3. Is it then possible to ensure that the return does not fall below 4 5%. the value at time 1 of a zero coupon bond with maturity 2 in this situation must be given by P1 2 = 1 ≈ 0 943 1 06 If interest goes down to 4%. In the following we show that the condition of no arbitrage only yields that 0 898 ≈ 0 962 0 943 <P 0 2 < ≈ 0 916 1 05 1 05 However.04) 1.8.06) 1 1. These observations are collected together in Figure 3. we are able to derive the price of many other contracts. we still cannot say precisely what the price P 0 2 at time 0 for the zero coupon bond with maturity 2 should be from the condition of no arbitrage alone. when the price at time 0 of the zero coupon bond is given. For example. when the interest in the lower part of Figure 3.113 (0. there are many different prices for the zero coupon bond which are consistent with the condition of no arbitrage.092 (0.05 (0.7 Arbitrage-free pricing in discrete time 81 1. .9. we obtain P1 2 = 1 ≈ 0 962 1 04 However.05 1.05) 1.113 1. assume that we know already at time 0 that we want to invest one unit at time 1.9 is 4%? We will introduce some additional concepts before we come back to this question.3. Model for the interest (numbers in parentheses) and the savings account S 0 . Actually.

The pair h t = h0 t h1 t is also called the portfolio at time t. 3.82 Interest rate theory in insurance 1 0.3 Investment strategies and the value process An investment strategy (or simply a strategy) is a two-dimensional process h = h0 h1 . The discounted value of the portfolio h t − 1 at . which describes when the number h1 of zero coupon bonds purchased and the deposit h0 on the savings account are chosen. The quantity h1 t represents the total number of units of asset number 1 which are in the portfolio at time t − 1.06) ? 1 (0.7. In particular. The condition concerning predictability of h1 ensures that the number of assets held from time t − 1 to time t is fixed at t − 1 based on the information available at this time. the insurance company’s portfolio is given by h t − 1 = h0 t − 1 h1 t − 1 i. This construction is illustrated in Figure 3. At time t − 1.962 1 (0. these assets have been a part of the investment portfolio from time t − 1 to time t.04) 1 Figure 3. Interest (numbers in parentheses) and the price for the zero coupon bond with maturity 2.9.943 (0.e. In contrast. h0 t is known/chosen at time t).e. such that the deposit at time t can be decided based on the additional information which arises from time t − 1 to time t. h1 t − 1 zero coupon bonds and a deposit on the savings account with value h0 t − 1 S 0 t − 1 . where h1 is predictable (i.e. the deposit h0 on the savings account is only required to be adapted. h1 t is known/chosen at time t − 1) and h0 is adapted (i.10.05) 0.

see also Föllmer and Schied (2002). We now turn to an analysis of the flow of capital. for applications in insurance.3. and thus the hedger receives the following discounted gains: h1 t X t − X t − 1 . Immediately after time t − 1. see Chapter 5 or Møller (2002) and references therein. Consider again the interval t − 1 t . This process can be applied for the construction of risk-minimizing strategies. see Figure 3. and this gives rise to the discounted costs h1 t − h1 t − 1 X t − 1 The new portfolio h0 t h1 t − 1 is held until time t. which takes place during the time interval t − 1 t . time t − 1. is defined by V t − 1 h = S0 t − 1 −1 h1 t − 1 S 1 t − 1 + h0 t − 1 S 0 t − 1 = h1 t − 1 X t − 1 + h0 t − 1 The process V t h t∈ 0 1 T is also called the (discounted) value process of h.10. when new prices S 0 t S 1 t are announced. The investment strategy h = h1 h0 and times for changes of investments.10. the portfolio h t − 1 is adjusted so that the hedger now holds h1 t bonds. For this purpose. This is achieved by buying an additional h1 t − h1 t − 1 bonds. where we use the savings account S 0 as the discount factor. We also refer to the insurance company as the hedger.7.4 The cost process Is is essential to describe changes in the value process and to keep track of whether changes are due to returns on investments or whether new capital has been added.7 Arbitrage-free pricing in discrete time 83 time: 0 1 t−1 t T savings account: h0(0) bonds: h1(1) h0(1) h1(2) h0(t − 1) h1(t) h0(t) h1(t + 1) h0(T) · Figure 3. 3. we introduce the so-called cost process suggested by Föllmer and Sondermann (1986) and Föllmer and Schweizer (1988).

Thus. 3. It decomposes changes in the value process into trading gains and changes in the accumulated cost process. if the portfolio is not affected by any in.48) The first and the last terms on the right hand side of Equation (3.or outflow of capital during the period considered. the hedger may at time t decide to change the deposit on the savings account from h0 t − 1 S 0 t to h0 t S 0 t based on the additional information available at time t. additional investments made during 0 t . i.50) .84 Interest rate theory in insurance Finally. given by t C t h =V t h − s=1 h1 s Xs (3. We now introduce the cost process (or the accumulated cost process) of the strategy h.5 Self-financing strategies and arbitrage A strategy is said to be self-financing if the change in the value process given by Equation (3. the cost process C h satisfies the following relation: V t h = V t − 1 h + h1 t X t − X t − 1 + C t h − C t − 1 h which corresponds to Equation (3.48). i.48) represent costs to the hedger. This change leads to the additional discounted costs h0 t −h0 t −1 . whereas the second term is trading gains obtained from the strategy h during t − 1 t . This condition amounts to requiring that t V t h =V 0 h + s=1 h1 s Xs (3. This means that any changes in the portfolio have to be made in a cost-neutral way. The cost process is simply defined as the value of the strategy reduced by trading gains.e.7. in the sense that the purchase of additional bonds must be financed by reducing the deposit on the savings account by a similar amount. in particular.48) is generated by trading gains only. We note that C 0 h = V 0 h . which says that the initial costs are exactly equal to the amount invested at time 0.e.49) where we have used the notation X s = X s − X s − 1 . we have seen that the change in value of the investment portfolio can be written as follows: V t h − V t − 1 h = h1 t − h1 t − 1 X t − 1 + h1 t X t − X t − 1 + h0 t − h0 t − 1 (3.

This characterizes the self-financing strategies in terms of the cost process.e. 3.7.51) The interpretation is exactly the same as in Example 3. In this case. With this strategy. In some situations (within certain models) it is possible to determine a self-financing strategy which generates or replicates the liability completely. one can invest zero units at time 0 and receive at time T the non-negative amount V T h . This is the case if there exists a self-financing strategy h which sets out with some amount V 0 h and has terminal value V T h = H at time T .7 Equivalent martingale measures and absence of arbitrage Typically we would like to verify that a given model does not allow for arbitrage possibilities. This fixes the distribution of the future prices and leads to a probability measure P. By inserting Equation (3.7. it corresponds to receiving a free lottery ticket. This is the case in the example considered in Section 3.3. i. which is strictly positive with positive probability. equal to the value of the initial investment made at time 0. Equivalent martingale measures When we specify how the price processes S 0 and S 1 develop.6 Hedging and attainability The insurer is assumed to invest on the financial market in order to control the risk associated with some liability H. and this leaves open the question of how to choose an optimal trading strategy. 2002). in many cases the hedger’s liabilities cannot be hedged perfectly using a self-financing strategy. we see that this corresponds to saying that the cost process is constant and equal to V 0 h . For a theoretical discussion on the choice of trading strategy for insurance contracts with financial risk. we also have to associate probabilities to the event that prices attain any given value. However. An arbitrage is a self-financing strategy h such that V 0 h =0 P V T h ≥0 =1 and P V T h >0 >0 (3. the initial value V 0 h is the only reasonable price for the liability H. the insurer is hedging against some risk in order to control or eliminate the risk.49).7. the concept of equivalent martingale measures plays a central role. such claims are also said to be attainable.7 Arbitrage-free pricing in discrete time 85 for all t. Here.8. see Møller (2001a.. 3.e.50) into Equation (3. and V 0 h is called the arbitrage-free price of H. . i.11.

Loosely speaking. We can modify the question slightly and ask: what is the expected discounted price under Q for the bond at time u. say. the answer is E P X u . see Section A. If we again compare with the situation with two different mortality intensities. We write P A for the “probability of A under P. this means that the measures agree on whether an event is totally unlikely or not. then it means that the mortality intensity is P if we use the probability measure P and Q if we use the probability measure Q. This situation can. we only require that the two probability measures are equivalent. be compared with the way the actuary handles simultaneously various bases for the mortality of a policy holder. and if we use Q the answer is E Q X u . we specify the probabilities for a certain class of events A (for example.2.” where we have added “under P” in order to underline the specific probability and the fact that one could as well speak of the probability of A under another measure Q. It is essential to introduce another probability measure Q. which is the ratio between the price of a bond S 1 and the value S 0 of one unit invested at time 0 of the savings account. we also have that Q A = 0 (and vice versa).1). we briefly recall what a martingale is. Basically. we deal primarily with the development of bond prices over time. that tossing a coin leads to “head” as in Section 3. for example. We can now ask questions such as: what is the expected (discounted) price of the bond at time u? Here.86 Interest rate theory in insurance More generally.52) . the two measures P and Q are now equivalent if we have that P is zero. if and only if Q is zero. which associates to events probabilities which might differ from the ones under P. Before we explain why the concept “equivalent martingale measures” appears in the title of this section. In this part. it is possible to formalize this change of mortality intensity by exactly the same method as the one sketched below. which means that if P A = 0. one has to be a little more precise and specify the probability measure that one wishes to use for the calculations – P or Q? If we work with P. Actually. For more details.1 of the Appendix. given that we take into consideration the amount t of information available at time t < u? This quantity is EQ X u t (3. At the moment. What is a martingale? Let us again consider the discounted price process X. and the two different probability measures represent different expectations for the changes in bond prices for a given period.

in particular. i.7 Arbitrage-free pricing in discrete time 87 the conditional expected value under Q of X u given the information available at t. for any t < u ∈ 0 1 that E Q X u t = X t . i. i. the expected discounted price tomorrow (calculated under the probability measure Q) is identical to the price today. such that the discounted T .e. i.e.52) is exactly equal to the discounted price at time t. The process X is now said to be a martingale (under Q). EQ X u t =X t (3. i.3.51). This condition becomes slightly more transparent in the case where X is a Markov process.e. we have price process X is a Q-martingale. We want to show that there cannot exist any self-financing strategies h with V 0 h = 0 and where T V T h =V 0 h + s=1 h1 s Xs (3. However. Consider some equivalent martingale measure Q. Here we introduce the following assumption.e. if an event has probability zero. it determines the events that are completely unlikely. Equivalent martingale measures An equivalent probability measure Q is said to be an equivalent martingale measure if X is a martingale under Q. We emphasize that the discounted bond price X is typically not a martingale under the true probability measure P. The importance of the concept of equivalent martingale measures is illustrated below.53) then reduces to EQ X u X t = X t (3.55) . if Equation (3. Assumption 3.. so that P normally cannot be applied as an equivalent martingale measure. if Equation (3. and if Q is a martingale measure. since Equation (3. Martingale measures ensure absence of arbitrage It is actually not so difficult to show that the existence of an equivalent martingale measure is sufficient to exclude the possibility of arbitrage possibilities in the model.12 There exists at least one equivalent martingale measure. If X t and X u represent the discounted price at time t (today) and time u (tomorrow) of a bond. X t .53) for all t ≤ u.53) is satisfied for all t < u.e. the measure P is important since.54) We interpret this as follows. that there cannot be strategies which satisfy Equation (3.

88

Interest rate theory in insurance

is non-negative and strictly positive with probability 1. To see that such strategies cannot exist, it is sufficient to show that EQ V T h = V 0 h (3.56)

Why does this exclude arbitrage possibilities? Well, if V T h ≥ 0 and EQ V T h = V 0 h = 0 we must have that Q V T h = 0 = 1, and hence P V T h = 0 = 1, since P and Q are equivalent. We still need to show that Equation (3.56) is satisfied for any self-financing strategy h. Here, we need the special construction of the investment strategy h, which ensures that h1 is predictable, i.e. that the number h1 s of bonds in the portfolio during the period s − 1 s is known at time s − 1. Using this property, we obtain the following: E Q h1 s Xs s−1 = h1 s E Q Xs s−1 = 0

where we have moved h1 s outside the expected value. (Mathematically, we have used that h1 s is s − 1 -measurable.) The second equality follows by noting that X is a Q-martingale, so that EQ Xs s − 1 = EQ X s − X s − 1 s−1

= X s−1 −X s−1 = 0 In fact, this shows that the value process given in Equation (3.50) for a self-financing strategy is also a Q-martingale, i.e. EQ V u h t =V t h

for all t ≤ u. If we insert u = T and t = 0, we see that EQ V T h = V 0 h Thus, under Assumption 3.12 there exist no strategies h that satisfy Equation (3.51). Pricing with an equivalent martingale measure We now consider some liability which specifies the discounted payoff H, which can be replicated by using a self-financing strategy h, i.e.
T

H =V T h =V 0 h +
t=1

h1 t

Xt

(3.57)

3.7 Arbitrage-free pricing in discrete time

89

In Section 3.7.6 we introduced the notion of an arbitrage-free price and we explained that this price is equal to V 0 h , the price needed in order to generate H with the self-financing strategy. We can now use the martingale measure Q to compute expected values in Equation (3.57). Since the value process V h is a martingale under Q, we see that EQ H = EQ V T h = V 0 h (3.58)

This very important result shows us that the arbitrage-free price for the liability can actually be calculated as the expected value of the discounted payments, computed under the equivalent martingale measure Q.

3.7.8 The two-period example revisited
Let us return to the example considered in Section 3.7.2 and assume that we have observed the price P 0 2 at time 0 of the zero coupon bond with maturity 2. The other prices at times 1 and 2 have already been established; see Section 3.7.8. Do there exist martingale measure(s)? How can we determine a martingale measure? Define a probability measure Q via q = Q i 2 = 0 04 = 1 − Q i 2 = 0 06 where 0 ≤ q ≤ 1. One can now start to think about whether Q is equivalent to the original measure P, i.e. if the two measures agree about whether an event has probability 0 or a strictly positive probability. Under P, the probability of i 2 attaining the values 0.04 and 0.06 is strictly positive. Hence this should also be the case under Q; this observation gives the condition 0 < q < 1. Next, we check if it is possible to choose q in such a way that Q actually becomes a martingale measure. Here, we need to verify that the process Xt = Pt 2 S0 t

is indeed a martingale under Q. Thus, it should be shown that EQ Pu 2 S0 u t = Pt 2 S0 t (3.59)

for 0 ≤ t < u ≤ 2. First, consider the case t = 1 and u = 2, so that P u 2 = 1 and S 0 u = S 0 t 1 + i 2 . Since the interest i 2 is assumed to be known at time t = 1, Equation (3.59) implies that 1 =P 1 2 1+i 2

90

Interest rate theory in insurance

which is exactly the formula that we derived already in Section 3.7.2. For u = 1 and t = 0, Equation (3.59) states that EQ P1 2 S0 1 = P0 2 =P 0 2 S0 0 (3.60)

In principle, we should include here the information 0 available at time 0. However, in our model there is no (non-trivial) information available at time 0, so this can be neglected. In order to be able to calculate the expected value on the left hand side of Equation (3.60), we note that the probability (under Q) that i 2 is equal to 0.04 is q (in this case, P 1 2 = 0 962), and that the probability that i 2 is equal to 0.06 is 1 − q (in this case, P 1 2 = 0 943). Thus we obtain the following condition: P 0 2 = EQ P1 2 S0 1 = 1 q 0 962 + 1 − q 0 943 1 05

which has the unique solution q= 1 05P 0 2 − 0 943 0 962 − 0 943

In particular, this shows that if 0 943 < 1 05 P 0 2 < 0 962, then 0 < q < 1 and hence Q is indeed an equivalent martingale measure in this case. We see moreover that the martingale measure in this example is uniquely determined from the prices which are given on the market. We can say that the market determines the martingale measure uniquely in this case.

A very simple interest rate guarantee Consider now the example where we are interested in a protection against the situation where the interest in the second period falls below 4.5%, say. Assume more precisely that we know that we will invest one unit at time 1 (for example because we receive some insurance premiums at this time), and that we want a guarantee which helps us in the scenario where i 2 is 4%. We therefore consider the contract which pays at time 2 the amount max 1 045 1 + i 2 . The discounted payment is given by H= max 1 045 1 + i 2 S0 2 = 1 1 05 1 + i 2 1 + i 2 + 0 045 − i 2
+

(3.61)

3.7 Arbitrage-free pricing in discrete time

91

where + is the positive part. According to Equation (3.58), the arbitrage-free price at time 0 for this contract is as follows: EQ H = = 1 1 0 005 + q + 1−q 0 1 05 1 05 1 04 1 0 005 +q 1 05 1 04 · 1 05

We see that the price consists of two terms, where the second term is the price for the guarantee. This price depends on the price of the zero coupon bond; see Figure 3.11. We end this example by showing how the interest rate risk can alternatively be hedged by buying zero coupon bonds and by using the savings account. More precisely, we construct a self-financing strategy h with a terminal value V T h , which coincides with the guarantee part of Equation (3.61); see Equation (3.57). Since the interest rate i 2 is known at time 1, we can construct a figure similar to Figures 3.7 and 3.8 with prices for the interest rate guarantee. This is illustrated in Figure 3.12. A self-financing strategy which generates the guarantee part of the contract, Equation (3.61), must satisfy the following equation: V 0 h + h1 1 X 1 + h1 2 X2 = 1 1 05 1 + i 2 0 045 − i 2
+

In our simple example,

X 2 = 0 and X1 = P1 2 −P 0 2 1 05

price for interest rate guarantee

0.004 0.003 0.002 0.001 0.000 0.900 0.905 0.910 0.915 zero coupon bond price

Figure 3.11. The price q 0 005/1 04· 1 05 for the interest guarantee as a function of the price at time 0 of the zero coupon bond with maturity 2.

92

Interest rate theory in insurance

0 0

(0.06) q 0.005 1.04·1.05

0

(0.05)
0.005 1.04

0.005

(0.04)

0.005

Figure 3.12. Model for the interest i t (numbers in parentheses) and the price for the interest rate guarantee.

so that we need to determine h1 1 and h0 0 such that V 0 h + h1 1 P1 2 −P 0 2 1 05 = 1 1 05 1 + i 2 0 045 − i 2
+

This leads to two equations (one for each of the possible outcomes for the future interest): h0 0 + h1 1 P 0 2 + h1 1 h0 0 + h1 1 P 0 2 + h1 1 0 962 −P 0 2 1 05 0 943 −P 0 2 1 05 = 0 005 1 05 · 1 04

=0

These two equations are solved as follows: h1 1 = 0 005 ≈ 0 253 1 04 0 962 − 0 943 0 005 · 0 943 ≈ −0 227 1 05 · 1 04 0 962 − 0 943

h0 0 = −

The example shows that it is indeed possible to hedge the interest guarantee by buying 0.253 units of the zero coupon bond and by borrowing 0.227 units from the savings account. With this strategy, the insurer is protected against the drop in the interest rate. We point out that the price for the interest rate guarantee, see Figure 3.12, and the investment strategy are both independent

t r 0 d t is the information available at time t.64) The savings account S 0 is defined by letting S 0 0 = 1 and dS 0 t = r t S 0 t dt It is well known that S 0 is then given by S 0 t = exp As mentioned above. this means that W t ∼ N 0 t . . and that W is continuous. and inserting these into Equation (3. The measure P is only used to specify whether an event is completely unlikely (has probability 0) or not. simulating independent normally distributed random variables W t . We take t = W u u≤t which means that we observe the Brownian motion W . 3.62) for the computation of the increments for r. We assume that the change in spot rate during a small time interval t t + t can be approximated by r t+ t −r t = r t ≈ t rt t+ t rt W t (3. For a more detailed study of discrete-time bond markets. In particular.63) Similarly.8 Models for the spot rate in continuous time In this section we briefly describe some so-called diffusion models for the spot rate.8 Models for spot rate in continuous time 93 of the original probability measure P. we also use the following notation: dr t = t 0 t r t dt + t r t dW t (3.62) is only valid for very small (infinitesimal) time intervals. Typically this corresponds to observing the interest r. To underline that Equation (3. are known functions. which means that W has independent. We assume that the process W is a Brownian motion.62) Here. with W t − W s ∼ N 0 t −s (mean 0 and variance t −s).63) and write r in the following form: r t =r 0 + s r s ds + t s r s dW s 0 (3. we refer to Jarrow (1996).3. normally distributed increments. we have used the notation W t = W t + t − W t . we can integrate Equation (3. It is possible to simulate r by choosing some interval length t.

94 Interest rate theory in insurance In Section 3. prices are not determined by the probability measure P. the function describes how the interest rate evolves under the martingale measure Q.” With this martingale measure Q. Some classical examples of interest rate models of the form in Equation (3. We refer to Equation (3.63) and (3. In contrast.65) as the “dynamics for r under Q. Estimation of parameters The functions and .59) for the similar result in discrete time.65). and this function cannot be estimated by observing the interest rate alone. In particular.65) where the function has been replaced by another function . are typically specified in such a way that they depend on some parameters. Equation (3. Equation (3.66) for various choices of parameters. for example. The function appears in both Equations (3. These parameters can be estimated by observing the interest rate over a certain period and then finding the value which describes the observations as well as possible. and where W is a Brownian motion under Q.65) are the Vasiˇ ek model. This ensures that the discounted price process P t T /S 0 t is indeed a Q-martingale.67) (see Vasiˇ ek (1977)).7 we showed that prices of zero coupon bonds and derivatives could be calculated as an expected value under an equivalent martingale measure Q. for example via maximum likelihood estimation. we mentioned that the measure Q should be determined from the zero coupon bond prices that are given on the market. c c . and then find the parameters which give the best description of the observed prices. the price at time t of a zero coupon bond maturing at time T is given by S0 t S0 T T P t T = EQ t = E Q exp − r t d t (3. which describe the behavior of the spot rate under the probability measure P. We now consider some martingale measure Q and assume that dr t = t r t dt + t r t dW t (3. In addition. compute Equation (3. Here one needs to include some observed zero coupon bond prices.66) see.

8. .67) (3.68) (see Cox.8 Models for spot rate in continuous time 95 and the Cox–Ingersoll–Ross model. and for the Cox–Ingersoll–Ross model. this level is b. The equations can be solved by first determining the function B from the first equation and then inserting this solution in the second equation in order to find A. For the Vasiˇ ek model with parameterization.68). Equation (3.2) for a proof of this result): P t T = E Q exp − T r t d t = exp A t T − B t T r t (3. Ingersoll and Ross (1985)): dr t = b − ar t dt + dW t dr t = a b − r t dt + r t dW t (3. for fixed t. which allow for relatively simple formulas for zero coupon bond prices. Equation (3. These two models are examples of so-called affine models.67).69) where A and B solve the following equations: t t Bt T + At T = tBt T − 1 t Bt T 2 1 tBt T − t Bt T 2 2 = −1 2 with B T T = A T T = 0. Equation (3.3.) We have the following general result for the price of a zero coupon bond with maturity T (see. 3. (Thus. Björk (2004. for example. this level c is given by b/a. Proposition 17.1 Affine models What is an affine model? How can prices for zero coupon bonds be calculated in the Vasiˇ ek and Cox–Ingersoll–Ross models? Is it possible to determine c the instantaneous forward rates in these models? An interest rate model of the form given in Equation (3. t · and 2 t · are affine functions of r.65) is said to be affine if t rt t rt = = trt + trt + t t where and are known functions.68) Both models have the special property that the interest rate returns to a certain level.

whether the interest rate r t is above or below the level given by Equation (3.72) as the maturity goes to infinity.71) Bt T 4a Since e−a T −t → 0 as T → . it can be shown that f t T = r t e−a T −t + + e−a T −t 2 2 2 b − 2 a 2a 1 − e−a T −t (3. This is not the case for the Cox–Ingersoll–Ross model.67). we obtain by solving the differential equations for A and B the following: Bt T = At T = 1 1 − e−a T −t a B t T − T + t ab − 1/2 a2 2 − 2 Bt T 4a 2 The forward rates at time t can be determined by differentiating these expressions with respect to T and then inserting the results into Equation (3. a phenomenon which is most likely for large values of .70). This shows that the forward rates given by Equation (3. We see that the shape depends crucially on the level of the interest rate r t at the time considered. in particular. Figure 3. In this way. we see that B t T → 1/a. . the forward rates might become negative.72).29) from Definition 3. Similarly.71) converge to 2 b − 2 a 2a (3. The Vasiˇ ek c model has the rather unfortunate property that it allows for negative interest rates. which does not allow for negative interest rates. A couple of simulations for r can be found in Figure 3.96 Interest rate theory in insurance The instantaneous forward rates at time t can be found via Equation (3.14.5: f t T =− T log P t T =r t T Bt T − T At T (3.13 presents a few examples of the shape of the forward rate curve.70) The Vasiˇ ek model c With the model given in Equation (3.

= 0 03. 0.02 0. b/a = 0 06.08 0.00 0 (a) 0. Parameters used: a = 0 36. = 0 09.00 0 5 10 15 time 20 25 30 Figure 3. = 0 03. Two possible simulations for r in the Vasiˇ ek model with initial c interest rate r 0 = 0 03.06 0.02 0.06 0.08 0.8 Models for spot rate in continuous time 97 0. forward rate forward rate .14.10 0.00 0 (b) 5 10 15 maturity 20 25 30 5 10 15 maturity 20 25 30 Figure 3. b/a = 0 06.04 0. Parameters: a = 0 36.04 0.15 spot rate 0.13.05 0.3. (a) Low standard deviation. Forward rates in the Vasiˇ ek model with high initial interest c r 0 = 0 07 (solid curves) and low initial interest r 0 = 0 03 (dashed curves). (b) high standard deviation.

73) .7 and 3.8 we showed how the price at time t of a zero coupon bond with maturity n could be written in the following form: P t n = EQ S0 t S0 n t where the expected value is calculated by using some equivalent martingale measure Q.7 and 3.3. it is possible to determine the forward rates by using Equation (3.2.3 we presented methods for deriving the market value of the guaranteed payments in a model where the interest rate could be stochastic.68). This result can now be inserted into the formulas derived in the main example in Sections 3. In Sections 3.98 Interest rate theory in insurance The Cox–Ingersoll–Ross model With the model given in Equation (3.70). The market value at time u of the payments guaranteed at time t can hence be rewritten as follows: V g t u = ba t E Q + n u S0 u S0 n S0 u S0 s u u n−u px+u EQ s−u px+u x + s bad − ds (3.5. However. we obtain Bt T = At T = where = a2 + 2 2 2e +a e 2ab 2 T −t T −t −1 −1 +2 T −t log 2 e a+ 2 + a e T −t − 1 + 2 Again.2 and 3. These results were stated in terms of either prices of zero coupon bonds or forward rates. the formulas turn out to be considerably more complicated than the ones obtained for the Vasiˇ ek model. c 3.9 Market values in insurance revisited We end this chapter by illustrating how the market values of the guaranteed payments may be calculated as the expected value under an equivalent martingale measure of the discounted payments. In Sections 3.

Equation (3.75) This brings to the surface the consequences for the market value. .76) Here. of allowing for stochastic interest rates. the uncertainty related to the policy holder’s lifetime and the uncertainty related to the future development of the interest rate appear simultaneously. The only randomness in Equation (3. However. it is possible to work with the payment process B t s introduced in Chapter 2. which described the payments guaranteed at time t.74) as follows: V g t u = E Q ba t exp − + n u n s r u d r u d n−u px+u exp − s r u d s−u px+u x + s bad − ds u (3. N is a counting process which counts the number of deaths (and hence which is either zero or one) and s n = 1 if s ≥ n and zero otherwise.e. This process is defined by dB t s = − I s ds + bad dN s + ba t I s d s n where I s is the indicator for the event that the insured is alive at time s.75) can finally be rewritten as follows: V g t u = EQ n u exp − s r u d dB t s u (3.9 Market values in insurance revisited 99 which can be recast as V g t u = E Q ba t + n u S0 u S0 n n−u px+u S0 u S0 s s−u px+u x + s bad − ds u (3.24). the uncertainty related to the policy holder’s lifetime has been averaged out. In these situations. market values are calculated as expected values under the equivalent martingale measure. we can use the fact that S0 u = exp − S0 s to rewrite Equation (3.3. Equation (3. With this notation. i.75) is that related to the development of the interest rate.74) In models with a continuously added interest rate.

100 Interest rate theory in insurance The market value is calculated directly as an expected value under a martingale measure of the random variable n u exp − s r u d dB t s (3. for example stocks. it has been demonstrated that the market values can alternatively be computed by using equivalent martingale measures. Equation (3. The next chapter investigates how the market value of the total liability is affected by changes in the interest and in other assets. This chapter has focused on determining the market value for the guaranteed payments in situations where the interest rate is stochastic.77) is the natural starting point for a discussion of market values in life and pension insurance. Finally. we have shown how forward rates can be applied to derive market values and the individual bonus potential.77) From a theoretical point of view. Moreover. Such considerations typically rely on a specification of the bonus and investment policy of the insurance company. . In this situation. This analysis is continued in Chapter 6. it is natural to replace the usual discount factors by zero coupon bond prices.

In Chapter 3 we introduced a stochastic interest rate and a bond market and we discussed the consequences for valuation in general and for valuation of guaranteed payments in particular. namely the possibility of investing in the risk-free interest rate. the condition is that the undistributed reserve. we introduce the possibility of investing in stocks and study the total reserve including the reserve for guaranteed payments. In this chapter we again assume a deterministic interest rate. We consider the type of insurance where the surplus is accumulated in the technical reserve leading to an increasing pension sum. which consists of one investment possibility only. In Chapter 3 we introduced a more realistic stochastic model for the interest rate and we discussed the consequences for the entries in the market balance scheme. Here. Furthermore. under certain conditions. binomial and Black–Scholes 4. in particular for the market reserve for guaranteed payments and the individual bonus potential. In Section 4. The condition and its consequences are formalized and studied in Chapter 2. but. be calculated using a simple retrospective accumulation. at the termination of the contract equals zero. which is the total reserve minus the technical reserve.1 Introduction In Chapter 2 we discussed some aspects of valuation assuming only one possible investment with a deterministic interest rate.6 we comment on the combination of stochastic interest rates and investment in stocks.4 Bonus. The total reserve in connection with a life insurance contract can. One very simple situation in Chapter 2 was the financial market. in return. In this chapter. The condition is that the total reserve which has been accumulated at the termination of the contract equals the pension sum paid out. we introduce investment possibilities in so-called risky assets or stocks and 101 . this risk-free interest rate is assumed to be deterministic.

which describe the bonus option as a financial derivative within the terminology of financial mathematics. 1997).102 Bonus. What mainly distinguishes the abovementioned contributions from each other is the way in which the bonus option is formalized. The collective bonus potential is the difference between the total reserve and the larges of either the technical reserve or the market reserve for guaranteed payments. we go through some calculations for certain relevant formalizations. or in another way: precisely how is the value of assets reflected in the bonus that is paid out? This is a fundamental question in this chapter. The Black–Scholes model has been the preferred model for the financial market in literature connecting bonus in life insurance with financial theory. As in Chapter 2. 2002). in which dividends are used to increase the pension sum. 2000. we consider an endowment insurance paid by a level premium. Thus. Hansen and Miltersen (2002). Furthermore. we disregard the mortality risk by setting the mortality intensity to zero. within which the authors give a formalization of the special payment streams appearing in life insurance contracts with bonus and possibly other options such as the surrender option. The starting point is two classical stock price models in mathematical finance: the binomial model and the Black–Scholes model. The binomial model is the simplest non-trivial stock model one can imagine. Seminal references are Briys and de Varenne (1994. 2003). the collective bonus potential is indirectly determined by calculating the total reserve. This provides a simple framework for the discussion of the various difficulties arising from a market valuation of insurance liabilities. Further important contributions are Grosen and Jørgensen (1997. . The purpose of this chapter is to develop theoretically substantiated methods for the valuation of the bonus obligations under investment in stocks. The power of the Black–Scholes model is that. The authors of the Black–Scholes model were awarded the 1997 Nobel Prize in Economics. in spite of its complexity. Miltersen and Persson (2000. it has some simple and appealing qualities that provide unique answers to certain pricing problems. The concepts of financial mathematics are mathematically much more involved in the Black–Scholes model than in the binomial model. which deal with the issues on a broader level. A complete reference list should also contain Møller (2000) and Steffensen (2001). However. and also Bacinello (2001) should be mentioned. The Black–Scholes model is a classical stock price model based on normally distributed returns over small time intervals. This makes it a popular stock price model in practice. The starting point is typically a Black–Scholes market. binomial and Black–Scholes study their consequences for the total reserve. pricing in the binomial model is not much more than the solution of two equations with two unknowns.

2 Discrete-time insurance model In this section we consider a discrete-time insurance model and the dynamics of various quantities of relevance for valuation within this model. However. The same notation is used in discrete and continuous time.and continuous-time insurance models are closely connected.2 we describe the relevant quantities in the market balance scheme in a discrete. Hereby. the Black–Scholes equation and the Black–Scholes formula may play important roles in the valuation of life insurance contracts with bonus. The actuarial notation is limited to a discount factor v and a t-year annuity at . In Section 4. The model is very simple and not very realistic. by tactical investments. An arbitrage argument is the argument for the value of a claim that if it had any other value. to obtain risk-free gains beyond what is offered by the financial market. In Section 4.5 we show how the Black–Scholes model. which both appear with decorations depending on the underlying interest rate. With appropriate arguments and an assumption of diversified mortality risk.and continuous-time financial mathematics contain similarities. the relevant version will be obvious from the context.3 we describe the binomial model in general.4 the Black–Scholes model is described and the celebrated Black–Scholes equation and Black–Scholes formula are stated and partly derived.3.2 Discrete-time insurance model 103 We wish to deal with the financial aspects exclusively.4.time model. where we give a systematic introduction to the binomial model. they are introduced in both connections such that both similarities and differences stand out. We calculate the quantities for one formalization of the connection between the bonus and the stock market. 4. we start in this section by studying a relatively complicated example of a product which can be treated within the binomial model. Correspondingly. all formulas can be adapted to a situation including mortality. In Section 4. the perspectives of the binomial model are emphasized to readers who prefer . The notions of discrete. The text contains a number of repetitions of definitions and arguments. This emphasizes the cross-sectional similarities. In Section 4. We formalize the notion of arbitrage and other concepts from mathematical finance in Section 4. In Section 4. Nevertheless. the discrete. it would be possible.6 we discuss briefly some generalizations of the considered market models. but it shows a lot of the problems that appear when trying to valuate insurance claims by means of so-called arbitrage arguments.

We speak of the periods as years. A lot of the quantities relate to quantities in Chapter 2. binomial and Black–Scholes to think of finance in terms of life insurance and bonus. The technical reserve V ∗ and the second order basis that is made up by the second order interest rate r are connected by the following equation which accumulates the technical reserve: V∗ t = 1+r V ∗ t V ∗ t − 1 + 1 t<n (4. such that it appears as a purely financial product in discrete time and such that investments. to which we refer for an interpretation. The insurance contract that we consider in this section is a modification of the main example in Chapter 2. as follows: V ∗ t = b t v∗ n−t − a∗ n−t In Chapter 2 we introduced the total reserve U . Møller (2001a) shows some applications of the binomial model to unit-linked insurance.called binomial market introduced below.104 Bonus.1) 0 = We also speak of the second order interest rate as the bonus interest rate and a process of bonus interest rates r t t=1 n as a bonus strategy. on the other hand. The sum b t guaranteed at time t is determined for a given technical reserve V ∗ t by the equivalence relation.1) and (4. The technical reserve and the first order basis which is made up by the first order interest rate r ∗ are connected by the following equation which accumulates the technical reserve: V ∗ t = 1 + r ∗ V ∗ t − 1 + 1 t<n + V∗ 0 = where the dividend payment t is given by t − r∗ V ∗ t − 1 t (4. take place in a so. The insurance sum paid out at time n is paid for by an annuity-due with premium payment .2) t = r such that the deposits stemming from Equations (4.2) coincide. which is obtained by an accumulation by the real basis: U t = 1 + r U t − 1 + 1 t<n U 0 = . but they could cover other time intervals.

4. The most simple asset with risk is an asset which. The total reserve U at time t is the value U at time t − 1 plus premiums plus capital gains. The idea now is to generalize this construction such that we have the opportunity to invest part of the money in an asset with risk. The company cannot wait until time t. The capital gains consist of interest earned on the part of U which was not invested in the risky asset at time t − 1. t must be decided at time t − 1. we know which investment gains we have in prospect the following year. This value is not known until time t. U t − 1 − t S t − 1 . which we also require from our investments later on. at time t − 1. experience the new asset price and then decide how many assets to buy at time t − 1.2 Discrete-time insurance model 105 Here. we let r be constant and. given its value at time t − 1. at time t takes one of two values but such that the value at time t is not known at time t − 1. We let S t denote the price of the risky asset at time t and let the dynamics of S be given by S t = 1+Z t S t−1 Here. Thus. Z t is a stochastic variable which takes the value u (for up) with the probability p u and the value d (for down) with the probability p d . and the change of the price of the portfolio of risky assets which follows from changes in asset prices. We assume that r ≥ r ∗ . Thus. the value X at time t is the value of X at time t − 1 plus capital gains plus the surplus contribution c t minus the dividend payment t . This is the intuitive content of the property predictable. known at time t −1. U t = U t−1 +r U t−1 − + t S t−1 t S t − S t − 1 + 1 t<n t S t − 1 Z t − r + 1 t<n = 1+r U t−1 + U 0 = The undistributed reserve X is given residually by X t = U t −V∗ t and we obtain the following: X t = 1+r X t−1 + where c t = r t − r∗ t V ∗ t − 1 Thus. in particular. Obviously. t S t−1 Z t −r +c t − t . We let the company buy t of these assets at time t − 1.

By fulfilment of the X n = 0. in the sense that the development in asset prices determine b n completely. but later we discover arguments for its construction.3). V g t . if we arrange the bonus strategy such that X n = 0. see Chapter 2. On the other hand. Very appropriately. in general.4) For now. is given by V g t = b t vn−t − an−t . The market value of the guaranteed payments. On one hand. in one way or another. we pull this probability measure from a hat. binomial and Black–Scholes In Chapter 2. If the bonus strategy is linked to the gains obtained by risky investments. the mathematics of finance concludes a modification of the prospective quantity as follows: V t = EtQ b n vn−t − an−t (4. The problem with this quantity is that. The following definitions are in correspondence with Equation (4. is linked to the development in asset prices. we need to define the market value of the payments guaranteed at time t. we also introduced the following prospective quantity for the total reserve and for the undistributed reserve: V t = b n vn−t − an−t This reflects the fact that the reserve is set aside to meet future obligations. Hereafter. then V t = U t . it requires a formalization of the future bonus strategy. happen with the following probabilities: r −d u−d u−r q d = u−d q u = (4. Finally. The introduction of risky investments makes the situation more difficult. we do not in general know b n at time t. if we assume that it is not linked to anything else. EtQ denotes the expectation given the development in the financial market until time t and under a very special probability measure Q. we can work with the retrospective quantities exclusively and disregard the future bonus strategy. and we want to avoid that. then this can be considered as a so-called contingent claim. no matter how complex the link may be. under which the outcomes that the risky asset goes up and down.3) Here. respectively. we wish that b n .106 Bonus.

4. the bonus strategy must obey the equivalence relation: V 0 = A natural question to ask here is whether there exists a situation such that the constraint X n = 0 again secures that the prospective quantities coincide with the retrospective quantities. the following: X n =0⇒V t =U t The implications following Equation (4.4). This can be seen since Q is exactly constructed such that EQ Z = r see Equation (4. We can write this using an inequality: U t d −Vg t d ≥ 0 − an−t (4. since the investment = 0 brings us back to the situation in Chapter 2. A sufficient condition on for X n = 0 is that we do not risk losing parts of V g via risky investments. less trivial than = 0.3). we give a sufficient condition on . obviously. In the following. we must ask ourselves whether there exists an investment strategy such that we can construct a bonus strategy leading to X n = 0? The answer is yes. for risk-free investments. Equation (4. also referred to as the bonus potential V b t . We have that X n =0⇒ b n = V∗ n = U n ⇒ V t = vn−t EtQ U n − an−t = vn−t v− n−t U t + an−t =U t In the third line we use the fact that the interest on U t and the premiums paid to U over t n under Q are expected to equal r. and the premium paid at time 0.5) generalize this result to the situation with risky investments.5) . In Chapter 2 we obtained. Let us check the consequence of the constraint. so saving us from a discussion on future bonus strategies. for a bonus strategy leading to X n = 0. Before we continue. no matter what the outcome of Z. is given by V b t = EtQ b n − b t vn−t In correspondence with the valuation formula.2 Discrete-time insurance model 107 The market value of the unguaranteed payments.

no matter that the outcome of Z. One can formalize a general investment strategy where the proportion of funds invested in risky assets. is a function of t V ∗ t − 1 X t − 1 . again. We can write this using an inequality: X t d ≥0 where. Then a natural question is: how is the bonus strategy linked to the financial market? One possibility is to suggest an appropriate explicit link which makes sense both from an economical and a mathematical point of view. If we add the classical solvency rule X t ≥ 0. X t = 1+r + +c t − t V∗ t−1 t X t−1 Z t −r X t−1 t V∗ t−1 X t−1 One can for example think of the special case where the value of investments in S is a constant proportion of U : t S t−1 = U t−1 ⇔ t V∗ t−1 X t−1 = = = t S t−1 X t−1 U t−1 X t−1 V∗ t−1 +X t−1 X t−1 If one disregards the condition X n = 0 in the market with risky assets. i. take as their starting points a certain explicit specification of the link and the authors then consider quantities like that gives in Equation (4. binomial and Black–Scholes where the argument d means that the outcome Z t = d is plugged in. t S t − 1 / X t − 1 . . introducing financial valuation in life insurance with bonus. If the constraint X n = 0 is fulfilled. the future bonus strategy is inevitably brought into the market values of unguaranteed payments. Many articles on the subject. a sufficient condition on is that we do not risk losing parts of V ∗ via risky investments.3). we have seen that V t = U t . t S t−1 = X t−1 Then.e. However. The conditions limit the investments in risky assets. one may wish to disregard these conditions.108 Bonus. the argument d means that the outcome Z t = d is plugged in.

1 we consider a two-period example with precisely this choice of function . given our simple financial market and one qualitative specification of the link between bonus and the financial market. that the bonus is set such that the part of the undistributed reserve which exceeds a constant buffer.e.6) where and are deterministic functions of t. Certainly. but just as important is to emphasize that there are no true or false answers to the above question. that the bonus is set as a constant part of the undistributed reserve if this is positive. there are no quantitative degrees of freedom if the bonus strategy has to be fair from a financial point of view. In Section 4.2 Discrete-time insurance model 109 It is important to discuss how the bonus strategy is currently decided. If t = 1 and t = −K < 0. is paid out.4. We take as our starting point the fact that the size of X determines the bonus strategy.2. there exist answers which do not conform with legislation but. legislation usually gives some degrees of freedom.e. We let the bonus interest rate be linked to X such that it is determined by r where Hereby. on the other hand. we obtain t X t−1 = X t−1 −K + i. We speak of as the contract function and note that a contract function in the form given by Equation (4. we obtain t X t−1 = X t−1 + i.6) contains two interesting constructions as special cases. K. t = r∗ t + t X t−1 V∗ t−1 is a function specifying how the bonus interest depends on X. . What we can show is that. c t − t = r t −r t V∗ t−1 t X t−1 V∗ t−1 V∗ t−1 = r t − r∗ t − =c t − One can think of a function t X t−1 in the following form: t + t x + t x = (4. If t = 0 and t = > 0.

we need two periods for an illustration: one period to earn capital gains and one period for the redistribution. Then we can write that r 2 becomes dependent on Z 1 but independent of Z 2 . whereas b 2 is independent of Z 2 .1. Because of this construction. we let the bonus interest rate in the second year depend on the capital gains in the first year. A construction where the capital gains are redistributed in the same year as they are earned could have been illustrated using a one-period model. We start out by giving in Figure 4. The consequence is that (see Figure 4. independently of the capital gains in that year but dependent on the capital gains in the preceding years. . We assume that the company at time 0 decides upon a bonus interest rate for the first year that does not depend on the capital gains in that year.1) V ∗ 1 and b 1 are independent of Z 1 . Hereafter. In that sense. However. the bonus interest rate is always “a year behind” the capital gains. Correspondingly.1 Two-period model In this section we consider a contract over two time periods and discuss bonus interest rates and investments. We let the bonus interest rate in the second year depend on the capital gains in the first year through the following formula: + X 1 V∗ 1 + r 2 = r∗ 2 + S(0) V ∗(0) U(0) b(0) V g(0) V b(0) S(1) V ∗(1) U(1) → b(1) V g(1) V b(1) V(1) → [V ∗(2)] = [b (2)] V(0) Figure 4. r 1 becomes independent of Z 1 . Hereby. binomial and Black–Scholes 4.1 an overview of the different quantities introduced in the previous section in the two periods. This accords with the conception of the bonus interest rate as an interest rate which is decided for the subsequent year.2. A two-period insurance model. we specify the connection between the bonus interest rate and the financial market. we let the bonus interest rate in the second year be independent of the capital gains in that year.110 Bonus.

the so-called bonus potential. r 2 u and r 2 d are the bonus interest rates in year 2 corresponding to the stock going up and down in year 1.7) . respectively. We determine this connection simply by the equivalence relation V 0 = which yields 1 1+r 2+r 2+r 1 1+q u r 2 EQ V ∗ 2 − 1 + EQ r 1+r 2 2 d = ⇔ 2 1 = 1+r + 1+r = 1+r 2+r ⇔ 2 u +q d r Here. we are interested in a connection between a fair investment and a fair pair of parameters in the bonus strategy . we derive the market value of the unguaranteed payments after the first year. If + X 1 d <0< + X 1 u (4. as follows: Vb 1 = = = = = = Q E1 V ∗ 2 − b 1 1+r 2+r 2+r 2+r 2+r 1 1 1 1 1+r 2 1+r 1+r 1+r −b 1 2 − V ∗ 1 1 + r∗ 2 1+r 2 − 2+r 1+r 1 1 + r∗ 2 r 2 − r∗ 2 1+r + X 1 + 1+r We see that this simple construction of bonus interest rates corresponds to a situation where the market value of unguaranteed payments after the first year is simply a part of the undistributed reserve. Hereafter.4.2 Discrete-time insurance model 111 and we obtain the following expression for the technical reserve after the second year: V∗ 2 = 2+r 1 1+r 2 which only depends on the capital gains in the first year. As a curiosity.

A redistribution parameter larger than unity makes sense since there is a bonus potential in the second year which must be redistributed but which is not included in X 1 . The risky asset can raise by 20% or fall by 10%. we can readily verify Equation (4. i. But with = 0. Before determination of and . In Figure 4. this is what we claim at this moment because we still have not presented the argument for the application of the artificial probability measure Q. u = 20%. Equation (4. We see that if our proportion of the risky asset is less than 55%. we consider a numerical example where we let s = 1 and = 1. With these numbers we find that q u = q d = 1/2. We let = 0 such that the bonus interest rate is determined exclusively by the parameter .8) = 1+r 2+r − 2+r Equation (4. and we then have to verify Equation (4. binomial and Black–Scholes we have that r 2 u > r ∗ = r 2 d . d = −10%. such that a bonus is paid out in the second year if and only if the risky asset goes up in the first year.7) is then obviously true.e. i.e. Since r 1 = r. This gives a redistribution parameter . but we shall come back to this later.7).7). the bonus interest rate in year 1 is also 5%. we have > 1. we determine the market values under a probability measure where the risky asset both rises and falls with probability 1/2.8) can determine fair combinations of the three parameters . The interest rate r is 5%.2 we find all quantities for the case where we have placed 60% in the risky asset. = 0 6. At least. i. we have that c 1 − 1 = 0 such that X 1 = s Z t − r . while the guaranteed interest rate is 3%.8) now gives a fair connection between and : = 0 55 Thus. Equation (4.e. In the rest of this section. we have that the redistribution parameter is a hyperbola as a function of the investment parameter . and furthermore we have + X 1 u 2+r 1 1 + r∗ + q u V∗ 1 = 1+r 2+r ⇔ 2+r 1 1 + r∗ + q u + r −r 1 + s/ 2+r 1 u−r = 1+r 2+r which implies that q u + r −r 1 + s u−r 1 1 + r∗ (4.112 Bonus.

4. the values above become natural as reserves for future obligations. independent of the outcome of the risky asset in that year.19 or 2. the bonus interest rate in the second year equals the first order interest rate. If the risky asset goes up in the first year. the bonus interest rate in the second year becomes 7%.11 V g(1) = 2. The argument simply shows how it is possible.09 S(0) = 1 V ∗(0) = 1 X(0) = 0 b(0) = 2. precisely to obtain the benefits that are due after the second year.2 V ∗(1) = 2.19] [b(2) = 2. .01 V b(1) = 0 V(1) = 2.05.01 [b(2) = 2.94 V b(0) = 0. In contrast.04 after the first year. both the individual and the collective bonus potential are 0.08 V(1) = 2.09 b(1) = 2.2. The difference between the bonus potential after the first year in the two scenarios pays the difference between the two benefits paid out after two years. all safety margins are lost and both bonus potentials are zero.11. Two-period insurance model.2 Discrete-time insurance model 113 S(1) = 1. We see that the technical reserve after the first year is 2. and thus the terminal payment.2 shows the outcomes of all quantities in Figure 4. equals 2.9 V ∗(1) = 2. If the risky asset goes down in the first year. 3%. We now give a brief argument for these values as they look in the example above.09 V g(0) = 0.06 V(0) = 1 S(1) = 0.1. of approximately 0. If the risky asset goes up in the first year.01 V b(1) = 0. depending on the outcome of the risky asset.91. Figure 4.11 V g(1) = 2. So far we have simply claimed how to calculate all values.05 X(1) = −0. by tactical investments of the premiums.11] Figure 4. the development of the technical reserve after the second year.05 X(1) = 0. Hereby.09 b(1) = 2. If the risky asset goes down in the first year. depending on whether the risky asset goes up or down in the first period.

no matter the outcome in the first year.44 and 1.114 Bonus. we shall not invest in the risky asset in the second year. Now we know what to do after the first year no matter the outcome in the first year. There are two cases corresponding to the risky asset going up or down in the first year. however. The number 1. but what should we do at time 0? Here again. no matter its outcome in the first. The solution is given by h1 2 u = 0 h0 2 u = 1 99 and we note that the price of this investment is precisely given by 1 99 × 1 05 = 2 09 = V 1 u A corresponding system is obtained if the asset goes down in the first year: h1 2 d 1 08 + h0 2 d 1 10 = 2 11 h1 2 d 0 81 + h0 2 d 1 10 = 2 11 which has the following solution: h1 2 d = 0 h0 2 d = 1 92 and the price for this investment is given by 1 92 × 1 05 = 2 01 = V 1 d As one would expect. Now. No matter what happens to the risky asset in the second year. since the bonus interest rate in the second year is independent of the outcome of the risky asset in that year.10 is 1 05 2 rounded off. we can set up a system where we are trying to hit what it takes to establish the required portfolios after the first year. which is the value after two years of one unit at time 0 including accumulated interest rate by 5%. the risky asset can either go up or down in the second year. Consider the case where the risky asset goes up.08 are the prices of the risky asset after the second year depending on the outcome of the risky asset in the second year and given that the asset goes up in the first year. that . We know. we need 2. If we let h1 2 u denote the number of risky assets and h0 2 u denote the investment in the risk-free interest rate.19 after the second year. binomial and Black–Scholes We calculate the investment which is required after the first year if the payment after the second year has to be met. we obtain the following equation system: h1 2 u 1 44 + h0 2 u 1 10 = 2 19 h1 2 u 1 08 + h0 2 u 1 10 = 2 19 where 1.

4.3.1 The one-period model In this section we consider the one-period version of the simplest possible nontrivial financial market. and we have thus constructed a portfolio strategy with which we can meet our obligations without taking any risk. but we must admit that they were constructed precisely such that the above calculations work out neatly. we used the probabilities q u and q d .3 The binomial model 115 in addition to the capital gains after the first year we also receive a premium payment of one. . The calculations boil down to two equations with two unknowns and. such that the equation system becomes h1 1 1 20 + h0 1 1 05 + 1 = 2 09 h1 1 0 90 + h0 1 1 05 + 1 = 2 01 This system has the following solution: h0 1 = 0 74 h1 1 = 0 26 and. we take a closer look at the construction of the measure Q and other concepts in financial mathematics.3 The binomial model 4. For the calculation of all quantities. finally. The arbitrage argument states that the market value of the obligation must be the price of establishing such a portfolio with which we can precisely meet the obligations. we mean that by calculating expected values with the probability measure Q. : 4. Hereby.2. we obtain values with which we can meet the obligations without risk. the model is often applied in practice. In the next section. Thus. The question is now: was this magic? The boring answer is as follows. we conclude the numbers in Figure 4. These seemed to be drawn from a hat. but because it is very simple to understand the concepts of financial mathematics in this market. where the stock after one period takes one value out of only two: not because we think that this market is very realistic. we see that this investment has exactly the value 0 74 × 1 + 0 26 × 1 = 1 = The price of this investment is exactly what is paid as a premium at time 0. furthermore.

we define a value process as follows: V t h = h0 S 0 t + h1 S 1 t t=0 1 The notion of arbitrage plays a very important role. in the portfolio at time 0. Linked to a portfolio h. and we define an arbitrage portfolio as a portfolio obeying the following relations: V 0 h =0 P V 1 h ≥0 =1 P V 1 h >0 >0 Thus. The interpretation of short-selling of bonds is that we borrow money in the bank at interest rate r. against a payment at time 0. but with a chance of being positive. We allow for negative components in the portfolio and interpret this as having sold the assets. The price of the stock is a stochastic process and takes the following values: S1 0 = s S1 1 = s 1 + Z Z= u with probability pu d with probability pd We study the dynamics of special investment portfolios in the market and define a portfolio (strategy) as a vector h = h0 h1 which describes the number of bonds and stocks. binomial and Black–Scholes The model consists of two investment possibilities: a bond and a stock. We can also interpret the investment opportunity S 0 as the possibility of saving money on a bank account with interest rate r.116 Bonus. and we then speak of the market as arbitrage-free. an arbitrage portfolio is a portfolio which is established at zero price at time 0. has a non-negative value at time 1. (4.9) . whereas the interpretation of short-selling of stocks is that we. respectively. for a given market one can pose the question. It is important here to emphasize that we usually require from a market that there exist no arbitrage portfolios. Such a position is called short-selling. Thus. oblige ourselves to pay the price of the stock in the future. The price of the bond is deterministic and takes the following values: S0 0 = 1 S0 1 = 1 + r S0 0 = 1 + r where r is a deterministic interest rate for the period 0 to 1.

A probability measure under which this is the case is called a martingale measure (or a risk-neutral measure or a risk-adjusted measure). A European call option gives the right. such a link between no arbitrage and the existence of a martingale measure goes far beyond the binomial model.11) If and only if we have the inequalities d < r < u. the following: 1 EQ S1 1 1+r = 1 q u s 1+u +q d s 1+d 1+r 1+q u u+q u d =s 1+r =s We see that the expectation of the discounted stock value at time 1 equals the stock price at time 0. Equivalently. valuation is a matter of what one can obtain. We assume that the market is arbitrage-free. d < r < u. From the reasoning above. in fact.e. It is important to note that whereas arbitrage is a matter of what one cannot obtain in the market. i. The function is called the contract function.4.10) (4. q d > 0 and the relation q u + q d = 1 in Equation (4. A contingent claim is a stochastic variable in the form X = Z . then q u . This is an important result and. Thus.3 The binomial model 117 is this market arbitrage-free? It is easy to see that the binomial market is arbitrage-free if d<r<u The solution q u q d to the following system of equations: q u u+q d d = r q u +q d = 1 becomes q u = r −d u−d u−r q d = u−d (4.11) assures us that we can interpret q u and q d as probabilities under a measure that we denote by Q. we conclude that a binomial model S 0 S 1 is arbitrage-free if and only if there exists a martingale measure. and we study the valuation problem for so-called contingent claims. it is determined by the outcome of S 1 1 exclusively.10). it is a stochastic variable since Z is a stochastic variable. but not the . A typical example of a contingent claim is a European call option on the stock with strike price K.12) (4. For this measure we have. according to Equation (4.

to buy a stock at time 1 at price K.15) d This is a system of two equations with two unknowns h0 h1 .118 Bonus. h is called a hedging or replicating portfolio. is determined as follows: h1 = h0 = u − d s u−d 1 1+u 1+r d − 1+d u−d u (4. the market is said to be complete. and the owner pays K for the stock and sells it on the market for s 1 + u with a gain equal to s 1 + u − K.13) Our problem is to find a fair price of this contract if one exists. i. A contingent claim is called attainable if there exists a portfolio h such that V 1 h =X (4. 0 X =V 0 h If the market is complete.14) In that case. there is in reality no difference between holding the claim and holding the hedging portfolio. the problem then is to find 0 X . If all contingent claims are attainable. X= Z = s 1+Z −K + = s 1+u −K Z = u 0 Z=d (4. under the assumption d < r < u. all claims can be priced by determining the price of the hedging portfolio. According to Equation (4.e. A natural question is now whether the one-period binomial market is complete. we know that for a portfolio hedging X. In our model this contract is only interesting if s 1 + d < K < s 1 + u .e. If S 1 1 < K. and we easily see that in order to avoid arbitrage we must have that 1 X = X.16) . the price of the claim at time 0 must equal the price of establishing a hedging portfolio at time 0. If a claim is attainable with the hedging portfolio h. the solution of which. The price at time t is denoted by t X . binomial and Black–Scholes obligation. Then it is easy to see that in order to avoid arbitrage possibilities in a market including the claim. the owner gives up the option with a gain of 0. If S 1 1 > K. h0 1 + r + h1 s 1 + u = h 1+r +h s 1+d = 0 1 u (4. i.14). the option is exercised.

d = −10%. as follows: 0 X = 1 EQ X 1+r where Q. After all.4. However. On the other hand.17) where q u and q d are given by Equation (4. K = 110. if we hold the option. with the parameters r = 5%. Risk neutrality is just a notion which shows up naturally when interpreting the quantities q u and q d as probabilities. We see that the price of a contingent claim X can be calculated by the so-called risk-neutral formula. The calculations above show that these two circumstances “level out” and that the disappearance of p u and p d does not seem that counter-intuitive after all.17) with q u q d given by Equation (4.3 The binomial model 0 X =V 0 h : 119 We can now calculate the price V 0 h = h0 + h1 s = = 1 1+u 1+r d − 1+d u−d d − 1+d u +q d u + u − d u−d u − 1+r d 1 1+u 1+r 1 = q u 1+r u + 1+r u−d d (4. given by the probabilities q u q d . p d = 0 4. it is important to realize that the pricing formula does not assume riskneutral agents. By risk neutrality we mean the intuitive quality that prices are calculated simply as expected discounted claims.12). the expected gain is affected by the objective probability measure. u = 20%. in which we find the price of a European option. The first idea could be to calculate the expected present value under the objective measure to obtain 1 P E 1 05 Z = 1 10p u + 0p d 1 05 = 5 71 . We conclude that the one-period binomial model is complete and that X = Z at time 0 has the price given by Equation (4. is the unique probability measure under which also the price of the risky asset can be calculated using a risk-neutral formula. We end this section with a numerical example.13). Equation (4. p u = 0 6. It may be surprising that the probabilities p u and p d do not play any important role in the pricing formula. by buying the option we give up the possibility of investing the price of the option on the same market with corresponding objective changes in prices.12).

Firstly. the risky asset and the option depend on the outcome of the stochastic variable Z. we write the price of the option at time 1.57 h1 = 0. we write S 0 and S 1 for the two time points 0 and 1.3. Hereafter.05 S1(1) = 90 ∏(1) = 0 Figure 4. which is given by Equation (4. We illustrate the completely different consequences of an arbitrage argument in Figure 4.05 S1(1) = 120 ∏(1) = 10 S0(0) = 1 S1(0) = 100 ∏(0) = 4.3. we calculate 0 and h from Equations (4.17) and (4.76 h0 = −28. the variance or the standard deviation.120 Bonus. binomial and Black–Scholes S0(1) = 1. Figure 4.33 S0(1) = 1.3 shows how the dynamics of the risk-free asset. for example. One-period option model. One may suggest loading that price according to. Finally.16).13). which in this example are as follows: 1 Q 0 = Z E 1 05 1 = q u u +q d d 1 05 1 0 05 − −0 1 0 2 − 0 05 = 10 + 0 = 4 76 1 05 0 2 − −0 1 0 2 − −0 1 h1 = h0 = 10 − 0 100 0 2 − −0 1 = 0 33 1 1 2 × 0 − 0 9 × 10 = −28 57 1 05 0 2 − −0 1 We can now verify that the price for establishing the portfolio h is exactly given by V 0 h = −28 57 + 0 33 × 100 = 4 76 whereas the value of the portfolio at time 1 becomes V 1 h = −28 57 × 1 05 + 0 33 × 120 = 10 Z = u −28 57 × 1 05 + 0 33 × 90 = 0 Z = d .

If we simply increase the number of outcomes and then search for replicating portfolios. is the portfolio held during the period t − 1 t . A portfolio (strategy) is a process h= h t t=1 n = h0 t h1 t t=1 n such that h t is a function of S 1 1 linked to the portfolio h is defined by S 1 t − 1 . It is intuitively clear that this portfolio only can depend on the prices until time t − 1 and not on the future prices from time t and onwards. the very simple asset model with only two possible outcomes. whereas the number of equations will equal the number of possible outcomes.15) will still be two. Thus. the completeness is maintained. the system will only have a solution for special trivial claims such as Z = S 1 1 .3 The binomial model 121 4.15) equals the number of unknowns. If we put a number of one-period models in a row. we allow agents to change their portfolio after each one-period model.3. the number of outcomes increases with the number of periods. respectively.4. h t . The value process V t h = h0 t S 0 t + h1 t S 1 t t=0 n The portfolio at time t. If. of course. We can say that the completeness of the binomial model exactly connects to the fact that the number of equations in the system shown in Equation (4.2 The multi-period model A drawback of the one-period model is. Otherwise. The answer is to allow for intermediary trading in the two assets. we get into trouble since the number of unknowns in Equation (4. on the other hand. h0 h1 . Particularly . We say that the portfolio must be predictable. we would of course take advantage of the future price changes when deciding the portfolio today. We assume a time horizon n and define the price processes for the risk-free asset and the risky asset as follows: S0 t = 1 + r S0 t − 1